Engineering Modeling, Analysis and Optimal Design of ...
Transcript of Engineering Modeling, Analysis and Optimal Design of ...
University of Massachusetts Amherst University of Massachusetts Amherst
ScholarWorks@UMass Amherst ScholarWorks@UMass Amherst
Open Access Dissertations
9-2011
Engineering Modeling, Analysis and Optimal Design of Custom Engineering Modeling, Analysis and Optimal Design of Custom
Foot Orthotic Foot Orthotic
Lieselle Enid Trinidad University of Massachusetts Amherst
Follow this and additional works at: https://scholarworks.umass.edu/open_access_dissertations
Part of the Mechanical Engineering Commons
Recommended Citation Recommended Citation Trinidad, Lieselle Enid, "Engineering Modeling, Analysis and Optimal Design of Custom Foot Orthotic" (2011). Open Access Dissertations. 498. https://scholarworks.umass.edu/open_access_dissertations/498
This Open Access Dissertation is brought to you for free and open access by ScholarWorks@UMass Amherst. It has been accepted for inclusion in Open Access Dissertations by an authorized administrator of ScholarWorks@UMass Amherst. For more information, please contact [email protected].
ENGINEERING MODELING, ANALYSIS AND OPTIMAL DESIGN OF
CUSTOM FOOT ORTHOTIC
A Dissertation Presented
by
LIESELLE E. TRINIDAD
Submitted to the Graduate School of the
University of Massachusetts Amherst in fulfillment
of the requirements for the degree of
DOCTOR OF PHILOSOPHY
September 2011
Mechanical Engineering
© Copyright by Lieselle E. Trinidad 2011
All Rights Reserved
ENGINEERING MODELING, ANALYSIS AND OPTIMAL DESIGN OF
CUSTOM FOOT ORTHOTIC
A Dissertation Presented
by
Lieselle E. Trinidad
Approved as to style and content by:
_______________________________________
Professor Sundar Krishnamurty, Chair
_______________________________________
Professor Ian Grosse, Member
_______________________________________
Professor Joseph Hamill, Member
____________________________________
Professor Donald Fisher, Department Head
Mechanical & Industrial Engineering
Department
DEDICATION
This dissertation is dedicated to my parents Juan and America Trinidad, without
their endless support I would have never been able to finish. Thank You!
v
ACKNOWLEDGMENTS
I would like to acknowledge my committee Dr.’s Sundar Krishnamurty, Ian Grosse and
Joseph Hamill for guiding me through this process. Dr. Krishnamurty, you supported me
for the past 7 years in my quest for “my own research project” I know that a large portion
of this research was not in your area of interest, I thank you for sticking with me. Dr.
Grosse, thank you for your assistance with the advanced knowledge on finite element
analysis which allowed me to proceed with this research. His acceptance to serve as my
committee member is highly appreciated. Dr. Hamill, you were always so warm, open
and willing to help me in any aspect of my research, even if it wasn’t your specialty. You
gave me great advice on how to proceed and where to look for answers, thank you!
Dr. Sandra Petersen, you were always there to support me in every aspect of my
graduate career ensuring that I progressed and that I was able to succeed in finishing my
dissertation. Our mutual mentoring sessions were invaluable.
I would also like to thank my friends and labmates for their support and
constantly finding a way to keep a smile on my face, Each of you always found a way to
keep things light when tensions were high. Jay Breindel, Sarah Wood, Andrew LaPre,
Michael Berthaume, Brianna Tomboulian, and in particular Christine Dzailo and Krishna
Samavedan. I would also like to thank my older labmates from my earlier years in grad
school Tiefu Shao, Paul Witherell, Brian Mullen , Justin Rockwell. From the kinesiology
department, my other department, I would like to thank in particular Dr. Ryan Chang, Dr.
Graham Caldwell, Dr. Sandy Whittlesey, Damien Callahan, Rebecca Hasson and
Catherine Gariepy. For my friends I would like to specifically thank Shelly Perdomo,
Millicent Jackson, Caryl Ann Becerra, Mckinley Milton, Radameris Gomez, Kyle
Morrison, Laura Hutchinson, Meaghan Germain, Khadine Higgins, Allison Guley,
Darlan Harris and Yvette Quinteros Dr. Vanessa Rivera, Jaime Chernoff and Dr.Ticora
Jones.
My coaches Krista shaus, Renee Willis and Gino Arcaro, having all of you to
manage my physical and mental helth over the last two years of this process has been the
last piece to put this puzzle together. Gino I know I only met you in the last month of
this process, but your words are what powered me through the finish line.
And finally I could not have done any of this without the support of my family, in
particular my father Juan Trinidad, my mother America Trinidad, my older sister Lenina
Trinidad and her husband Steve Kenny, my younger sister Ayla, My uncle and Godfather
Jose Acevedo and my Aunt Angela Trinidad. I would also like to thank my aunt and
Godmother Maria Trinidad, my aunt Milagros Castro, and the rest of my tribe of aunts,
uncles, cousins and grandparents.
vi
ABSTRACT
ENGINEERING MODELING, ANALYSIS AND OPTIMAL DESIGN OF
CUSTOM FOOT ORTHOTIC
September 2011
LIESELLE E. TRINIDAD, B.S., S.U.N.Y. BUFFALO
M.S.M.E, UNIVERSITY OF MASSACHUSETTS AMHERST
Ph.D., UNIVERSITY OF MASSACHUSETTS AMHERST
Directed by: Professor Sundar Krishnamurty
This research details a procedure for the systematic design of custom foot
orthotics based on simulation models and their validation through experimental and
clinical studies. These models may ultimately be able to replace the use of empirical
tables for designing custom foot orthotics and enable optimal design thicknesses based on
the body weight and activities of end-users. Similarly, they may facilitate effortless
simulation of various orthotic and loading conditions, changes in material properties, and
foot deformities by simply altering model parameters. Finally, these models and the
corresponding results may also form the basis for subsequent design of a new generation
of custom foot orthotics.
Two studies were carried out, the first involving a methodical approach to
development of engineering analysis models using the FEA technique. Subsequently, for
model verification and validation purposes, detailed investigations were executed through
experimental and clinical studies. The results were within 15% difference for the
experimental studies and 26% for the clinical studies, and most of the probability values
were greater than α = 0.05 accepting our null hypothesis that the FEA model data versus
vii
clinical trial data are not significantly different. The accuracy of the FEA model was
further enhanced when the uniform loading condition was replaced with a more realistic
pressure distribution of 70% of the weight in the heel and the rest in the front portion of
the orthotic. This alteration brought the values down to within 22% difference of the
clinical studies, with the P-values once again showed no significant difference between
the modified FEA model and the clinical studies for most of the scenarios.
The second study dealt with the development of surrogate models from FEA
results, which can then be used in lieu of the computationally intensive FEA-based
analysis models in the engineering design of CFO. Four techniques were studied,
including the second-order polynomial response surface, Kriging, non-parametric
regression and neural networking. All four techniques were found to be computationally
efficient with an average of over 200% savings in time, and the Kriging technique was
found to be the most accurate with an average % difference of below 0.30 for each of the
loading conditions (light, medium and heavy).
The two studies clearly indicate that engineering modeling, analysis and design
using FEA techniques coupled with surrogate modeling methods offer a consistent,
accurate and reliable alternative to empirical clinical studies. This powerful alternative
simulation-based design framework can be a viable and valuable tool in the custom
design of orthotics based on an individual’s unique needs and foot characteristics. With
these capabilities, the CFO prescriber would be able to design and develop the best-fit
CFO with the optimal design characteristics for each individual customer without relying
upon extensive and expensive trial and error ad hoc approaches. Such a model could also
facilitate the inspection of robustness of resulting designs, as well as enable visual
viii
inspection of the impact of even small changes on the overall performance of the CFO.
By adding the results from these studies to the CFO community, the prescription
process may become more efficient and therefore more affordable and accessible to
all populations and groups.
ix
TABLE OF CONTENTS
Page
ACKNOWLEDGMENTS ...................................................................................................v
ABSTRACT ....................................................................................................................... vi
LIST OF TABLES ............................................................................................................ xii
LIST OF FIGURES ......................................................................................................... xiv
CHAPTER
1. INTRODUCTION ...................................................................................................1
Statement of the Problem .........................................................................................6 Specific Aims ...........................................................................................................6
Study 1 .........................................................................................................6
Specific Aims #1: .............................................................................6
Study 2 .........................................................................................................8
Specific Aims #2: .............................................................................8
Significance of the studies .......................................................................................8 Summary ..................................................................................................................9 Previous work ........................................................................................................10
FEA model of a Custom Foot Orthotic ..........................................10
Introduction ........................................................................10
Methods..............................................................................11 2. LITERATURE REVIEW ......................................................................................15
Introduction ............................................................................................................15
Orthotics .................................................................................................................15
Orthotics & Finite Element ........................................................................17
Ankle Foot Orthotic Finite Element Analysis Research ............................18 Accommodative Orthotic Finite Element Analysis Research ...................19 State of the Art ...........................................................................................22
Metamodeling ........................................................................................................22
x
Summary ................................................................................................................24
3. METHODOLOGY ................................................................................................26
General Introduction ..............................................................................................26 Study 1 ...................................................................................................................26
Introduction ................................................................................................26 Verification and Validation........................................................................27
Experimental set-up and stress analysis .........................................27 Clinical trial and data collection ....................................................28
Subjects ..............................................................................31 Experimental set-up ...........................................................31
Data reduction ....................................................................32
Cantilever beam test .......................................................................32 Varying load location and arch height ...........................................34
Summary ....................................................................................................35
Study 2 ...................................................................................................................35
Introduction ................................................................................................35 Experimental set-up ...................................................................................36
Validation ...................................................................................................37
Model Building: .............................................................................38
Summary ....................................................................................................39
Summary ................................................................................................................40
4. RESULTS ..............................................................................................................41
Study I ........................................................................................................41
Instron Verification ........................................................................41
Clinical Validation .........................................................................43
Uniform Surface load .........................................................43 Varying distribution load ...................................................44
xi
Study II.......................................................................................................47
Model Validation: ..........................................................................54
5. DISCUSSION & CONCLUSIONS .......................................................................60
Study I ....................................................................................................................60
Conclusion .....................................................................................65
Case study: varying arch height and load location ....................................66
Study II...................................................................................................................69
Case study: Force load distribution using surrogate modeling ..................72
6. SUMMARY & FUTURE WORK .........................................................................81
Summary ................................................................................................................81
Future work ............................................................................................................82
APPENDIX: CLINICAL TRIAL DOCUMENTATION ..................................................83
BIBLIOGRAPHY ..............................................................................................................92
xii
LIST OF TABLES
Table Page
Table 1. Sample weight to thickness ratio guideline for orthotics prescribers .................36
Table 2. Data used to build the surrogate models for each scenario ................................39
Table 3. Deflection for Instron experimental tests vs. ANSYS WB deflection
values. ....................................................................................................................42
Table 4. Comparison of clinical trials data vs. ANSYS WB uniform load
deflection values. ...................................................................................................44
Table 5. Comparison of clinical trials data vs. ANSYS WB varying distributed
load deflection values 30/70. Light = 0.040MPa (88lbs), Medium =
0.074MPa (160lbs), Heavy = 0.138MPa (300lbs) .................................................46
Table 6. Comparison of clinical trials data vs. ANSYS WB varying distributed
load deflection values 60/40. Light = 0.040MPa (88lbs), Medium =
0.074MPa (160lbs), Heavy = 0.138MPa (300lbs) .................................................47
Table 7. Light scenario surrogate model compared to finite element model values
and percent difference for the Response Surface, Kriging, Non Parametric
Regression, and Neural Networking methods. ......................................................58
Table 8. Medium scenario surrogate model compared to finite element model
values and percent difference for the Response Surface, Kriging, Non
Parametric Regression, and Neural networking methods. .....................................58
Table 9. Heavy scenario surrogate model compared to finite element model
values and percent difference for the Response Surface, Kriging, Non
Parametric Regression, and Neural networking methods. .....................................59
Table 10. Clinical, uniform load and distributed load deflection values compared
using standard error and P-values ..........................................................................64
Table 11. Results from uniform and 30/70 simulations compared to clinical
results. The deformations are in negative z direction. The number in
parenthesis under the clinical deflection are standard deviation. ..........................73
Table 12. Weight Distribution Regional Mean Values and SD (N=107)
(Cavanaugh, 1987). ................................................................................................75
Table 13. Pressure distribution (%) for various loading conditions with
corresponding deformation Mean (SD) of N=10 ...................................................76
xiii
Table 14. Optimal Design Variable Pressure Distributions and corresponding
deformation values from various Goal Driven Optimization Techniques .............78
Table 15. Uniform, 30/70 and Optimal Pressure FEA results compared to clinical
data (the deformations are in the negative z-direction) .........................................79
Table 16. Optimal Plantar pressure Distribution for medium subject..............................80
xiv
LIST OF FIGURES
Figure Page
Figure 1. Custom foot orthotics manufacturing process flowchart ....................................3
Figure 2. a) Imbalanced spine due to foot pronation (or a low arch), b) balanced
spine .........................................................................................................................4
Figure 3. Four layers of a semi-rigid style custom foot orthotic ........................................5
Figure 4. Boundary conditions applied to FEA model simulating the midstance
phase of gait: (a) Zero DOF in the vertical direction on the bottom area
(arch left unconstrained), (b) Zero DOF in the underside of the heel region
in all directions, (c) Zero DOF in the horizontal direction on the back,
lateral and front edge, (d) uniform surface pressure load applied to the
entire top surface. ...................................................................................................13
Figure 5. Maximum deflection vs. the applied load by orthotic thickness. .....................14
Figure 6. Stress distribution view (a) top and (b) bottom. ...............................................14
Figure 7. (a) ankle-foot orthtotic, (b) accomodative orthotic ..........................................18
Figure 8. Experimental set-up for orthotic in Instron machine in MIE materials
laboratory ...............................................................................................................28
Figure 9. Uniformly distributed surface load application ................................................29
Figure 10. Orthotic used for clinical trials including tracking markers ...........................30
Figure 11. Cantilever beam material set-up including 1”x 3” rectangular bar of
polypropylene. A strain gage was glued onto the polypropylene bar and a
5N load was applied at a distance of 56mm from the clamp .................................33
Figure 12. ANSYS WB model, (a) Medial side view showing arch alteration
lines, (b) top view showing Instron 15mm diameter load location and
alteration lines as well as 30mm diameter constraints. ..........................................34
Figure 13. Split load application (30% of load applied to back/heel portion of the
model) ....................................................................................................................45
Figure 14. Surrogate models for the light load scenario (A) Response Surface,
(B) Kriging, (C) Non Parametric Regression, (D) Neural networking. .................49
xv
Figure 15. Surrogate models for the medium load scenario (A) Response
Surface, (B) Kriging, (C) Non Parametric Regression, (D) Neural
networking. ............................................................................................................51
Figure 16. Surrogate models for the heavy load scenario (A) Response Surface,
(B) Kriging, (C) Non Parametric Regression, (D) Neural networking. .................53
Figure 17. Two-Variable Face-Centered CCD ................................................................55
Figure 18. Two-Variable Rotatable CCD ........................................................................56
Figure 19. Two-Variable Inscribed CCD .........................................................................57
Figure 20. Effect of load location holding thickness and load constant separated
by arch ht. Only 3mm used for example, all thicknesses are similar. (A)
is the 60N load, (B) is the 140N load and (C) is the 200N load. ...........................68
Figure 21. Varying arch height (only 3mm is shown, other orthotic thicknesses
are similar) .............................................................................................................69
Figure 22. The 10 anatomical regions that result from regional division
(Cavanagh, 1987). ..................................................................................................74
Figure 23. (a) ANSYS workbench CFO model showing midline division, (b)
ANSYS workbench CFO model displaying imprinted faces for
redistributed plantar pressure distribution. ............................................................76
1
CHAPTER 1
INTRODUCTION
Today’s competitive manufacturing environment has placed a great importance
on the ability to reduce the time, effort, and expense pertaining to the iterative decision-
making process used in design of products. Engineering design typically requires the
collaboration of resources from mathematics, science, engineering and technology to
optimally convert resources to meet the desired needs (BMED, 2001). The quality of
engineering design and the inherent cost of ensuring such quality are two critical
attributes which will always be conflicting with each other. In general, a higher quality
design outcome typically requires higher developmental costs. Therefore, it is critical for
engineering decision makers to find the best method of minimizing their developmental
cost while still maintaining the highest quality possible. In addition, information
technology in conjunction with global free trade policies, encourage a significantly large
increase in competition from all over the world (Haythornthwaite, 2006, Goldman, 1999).
This surge of competition logically brings about price wars in which profit margins may
decrease all the way to a negative point if the product developing costs are not
sufficiently low (Curedale, 2003). When developing a product, over 70% of life-cycle
costs are expended during the design phase (Shao, 2007, Anderson, 2003). The quality of
the design outcomes has the largest impact on the engineering decision maker.
Engineering designs are driven by customers’ needs, yet constrained by their
buying power and the small window of opportunity to enter the market before unknown
competitors flood the market (Michalewicz, 2004). Therefore, a constant demand is
placed on engineering decision makers to be more efficient and productive in producing
2
superior designs (Blunkett & Johnson, 2005). “This requires a results-driven, effective
and efficient engineering design framework that facilitates the finding of optimal design
solutions at the minimum cost.” (Shao, 2007, Sobieszczanski, 1997).
Proper knowledge and understanding of how body forces are applied and the
mechanics of the interaction between the body and orthotic can facilitate the development
of optimally designed orthotics. Significant advancements in the understanding of
kinematics and dynamics of human motion, as well as in the design and development of
medical devices to enhance human performance, can offer new paradigms for the holistic
solutions to the challenges faced when quality of life is compromised. On the basis of
these considerations, this research aims to demonstrate how engineering modeling,
analysis and optimization can significantly enhance the design and development of the
Custom Foot Orthotic (CFO), a widely used performance enhancement device.
With over 50 CFO manufacturers and over half of North Americans in need of
orthotic intervention (Christopher Maclean, former president of PFOLA and director of
Biomechanics at Paris Orthotics), there is still little scientific evidence in the literature to
support the positive results seen by patients throughout the years, due to the effects of
stress redistribution and accurate foot shape support. This research offers
recommendations of methods to use that will further our understanding of the effects that
CFOs have on human movement and performance.
3
The current CFO manufacturing process is described in Figure 1. It is evident
that there are many steps in this process, which makes it a timely and expensive process.
Incorporating engineering modeling will allow increased efficiency of the manufacturing
process.
Figure 1: Custom foot orthotics manufacturing process flowchart (Image credits in order of appearance to: No Aestetic [http://sal2009.com/index.php?key=Orthotics], Podiatrytoday.com, ProLab
Orthotics [http://www.prolaborthotics.com/Education/Casting/CastEvaluationFunctional/tabid/226/Default.aspx], direct industrial [http://www.directindustry.com/prod/cnc-step-hylewicz-cnc-technik/3d-laserscanner-64456-442366.html],
http://www.oguiadacidade.com.br/video/easycad/, Thermwood blog [http://blog.thermwood.com/?Tag=5%20Axis], No Aestetic.)
Many people who are in need of orthotic intervention either cannot afford
them or are not willing to pay the price for a treatment that does not have concrete
positive claims. (Trinidad, 2008) People from varying communities use orthotics,
from frail elders to athletes and everyone in between. Orthotics are typically accepted
as a method for resolving ailments by altering the position of the foot, which in turn
alters the lower extremities and one’s alignment all the way up the body. Figure 2
4
demonstrates how a spine can become imbalanced by the pronation of the left foot.
This pronation causes the leg to shorten which causes the hip to tilt which in turn
causes the opposite shoulder to dip.
Figure 2: a) Imbalanced spine due to foot pronation (or a low arch), b) balanced spine (image credit to: Gibsons Chiropractic Clinic and Marotta Health and Wellness chiropractic)
The formal definition of an orthotic is “a support, brace, or splint used to support,
align, correct or prevent the function of moveable parts of the body” (Medicinenet, 2007).
Shoe inserts are orthotics that are intended to “correct an abnormal, or irregular walking
pattern, by altering slightly the angles at which the foot strikes a walking or running
surface” (Medicinenet, 2007). Over the last ten years, there has been an incredible
increase in the use of CFO’s as evident by the increased volume of foot orthotics
manufactured and the number of new orthotic labs (Richie, 2006). With the increase in
production, CFO prescribers want more knowledge about orthotic therapy, but one
obstacle is the lack of uniform information (Richie, 2006). Another is the high cost of the
is corrected Foundation When
5
intervention, without clear concrete evidence guaranteeing positive results; a typical pair
of CFOs ranges from three to five hundred dollars (Feldman, 2010).
Figure 3: Four layers of a semi-rigid style custom foot orthotic (Image credit to: Kintec FootLabs)
A semi-rigid style orthotic is a functional foot orthotic “used to partially control
abnormal motion or abnormal position of the foot and leg during gait”
(www.PFOLA.org/technicaltopics). The most popular materials used to make up a
typical semi-rigid style orthotic are polypropylene, EVA foam, Spenco™, Topy, and
McPuff (Figure 3). For the support material, polypropylene and graphite composite
comprise 98% of the CFO that are made (Richie, 2006). For the purposes of this research
we will not be discussing graphite composite.
Presently, clinicians calculate the stiffness of the orthotic based on their
experience using the patient’s characteristics (i.e. arch height, foot and body mechanics,
weight and activity level) and the orthotic design (selected material: orthotic shell, top
cover, posting material, alterations to the cast, and additions to the orthotic). In the
future, it seems possible to produce interactive software wherein the clinician can
complement traditional clinical methods with engineering models so as to produce
orthotics that are optimal for each particular client. Towards this end, this research
investigates the application of engineering modeling, analysis and modeling methods in
6
the design of CFOs. As part of this research, analysis models using finite element
analysis methods will be developed and their results will be verified and validated based
on experimental and clinical studies. Surrogate models will be developed through the use
of Response Surface Methods and Kriging techniques to further enhance the
computational efficiency of the design process. The following section details the specific
aims of this research and their significances.
Statement of the Problem
Previous research (Trinidad, 2008) detailed a simulation-based design procedure
for the systematic design of CFOs. Findings showed that when properly employed, the
models have the potential to enhance the prescription process by complimenting current
practices. Further extending this simulation-based approach to include predictive models
will enable optimal design of CFOs based on an individual person’s body weight,
activities, loading conditions, foot functions, etc.
Specific Aims
Study 1
Specific Aims #1:
The specific aims of study 1 are to develop refined Finite Element Analysis
(FEA) simulation models to mimic CFO behaviors under loading and validate the results
through clinical and experimental results. These tasks are important because the FEA
model of the custom foot orthotic is the first of its kind. CFOs have not yet been studied
using FEA to date. These models provide a starting point for further research on CFO’s
using FEA. Verifying and validating these models will ensure that the model mimics real
orthotic behavior.
7
The tasks include: a) running a clinical trial to validate the model, b) running an
experimental trial to verify the model, and c) investigating the influence of arch height
and load location on arch deformation. The challenges involved in these tasks include
designing experimental and clinical studies to asses and validate FEA model assumptions
and criteria, and altering the FEA model to account for various arch heights and load
locations. The FEA model geometry was created using a laser scanned image of the
physical orthotic and converting the image to a CAD model. Altering the geometry will
involve transferring the model to Solidworks (a CAD software) and creating a program to
modify the geometry. When designing the experimental and clinical trials, it is difficult
to apply all of the assumptions made in the development of the model. For instance, the
model does not contain a floor meaning there is no friction involved. Although in the
clinical and experimental setting the friction between the floor (or base of fixture) and
orthotic is very small and assumed to be negligible, there is still a difference. Another
difference is the shape of the orthotic. The orthotic in the model was built from a sample
orthotic, whereas the orthotics used in the clinical trial are tailor made for each subject.
The difference in geometry can affect model variables including length of the device,
heel depth, and arch height. In the experiment the orthotic must be held down by a clamp
in two spots to prevent the orthotic from sliding and twisting. In the lab the orthotic is
double stick taped to the bottom of the subject’s foot, which will prevent the foot from
fully expanding and deflecting the arch completely. Although there are some challenges,
these studies will greatly enhance the current CFO research. FEA is a tool that can allow
access to experiments not necessarily possible in a clinical trial, and its tools have not
fully been utilized in the investigation of CFOs to date.
8
Study 2
Specific Aims #2:
The specific aims of study 2 are to develop surrogate models and design
procedures to augment currently used empirical tables in the CFO prescription process.
The empirical tables are a reference used by clinicians assigning orthotic thicknesses to
certain weighted individuals (Table 2 in methods section). This study is important
because simulation models are computationally costly. One can incorporate many
different variables in a surrogate model without re-running a costly FEA model.
The challenges include obtaining enough data to support the development of
accurate surrogate models as well as validating the surrogate models. Research tasks
include building the RSM, Kriging, non-parametric regression and neural network model
from the FEA data using thickness, deflection and arch height as a function of weight. It
is hypothesized that the surrogate models will accurately mimic the FEA model
simulations and the resulting response surface will produce visual representation of the
input and output variables to replace empirical tables currently used.
Significance of the studies
Although orthotics have become widely used and accepted as devices for the
prevention of and recovery from injuries, the design process continues to be based on
empirical means. A deeper understanding of the therapeutic effects of a CFO and its
design for optimal performance can be achieved through systematic simulation-based
engineering modeling and analysis studies.
9
There has been very little conclusive research done pertaining to the
biomechanical influence of custom foot orthotic intervention. The majority of the
clinical studies performed to date have resulted in conflicting results due to significant
limitations such as experimental design or subject selection (Maclean et al., 2006).
Research applied to many rehabilitation devices is inconclusive due to many
uncontrollable factors, as is the case with research relating to CFO’s. Some of these
factors include “the large variability of individual foot types, differing philosophies on
foot function and a large variety of materials as well as types of orthotics” (Feldman,
2010).
This research stipulates that biomechanics research can be improved when clinical
studies are coupled with rigorous engineering methodologies. This research expands on
the understanding of human movement and performance through modeling, simulation
and analysis. Accordingly, it is the goal of this research to apply engineering modeling
and analysis to better understand the therapeutic effects of orthotics and to give insight
into the best method of research suited for future work and practical application in this
area. Therefore, these simulation-based modeling, analysis and corresponding results
offer a promising new approach to optimal design and development of CFOs.
Summary
Although orthotics are accepted as an effective means of treating and preventing
injuries, it may take months for results to be seen because the prescription process is
primarily one of qualitative means. Patients may also hold back on purchasing orthotics
due to the high cost. In addition, patient compliance may be lower than desired due to a
lack of scientific basis for the claim to positive results.
10
This research aims to verify and validate the finite element model previously
created, replace empirical tables with design charts using surrogate models, and offer
greater freedom in experimental design as well as in the CFO prescription process.
By adding these results to the CFO community the design process may become more
efficient, and the product may therefore become more affordable and accessible to all
population and groups.
Previous work
FEA model of a Custom Foot Orthotic
Introduction
In previous research we presented an engineering approach to the modeling and
analysis of Custom Foot Orthotics (CFOs). The development of a methodical simulation-
based approach using Finite Element Analysis (FEA) is outlined. Salient steps for the
development of accurate CFO models include the creation of FEA models through
approximation of complicated material properties, as well as a methodical process for the
replication of the complex three-dimensional geometry.
Despite the common practice of modeling and analysis through FEA in product
design, tools related to these systems have not been utilized when designing custom
prescription foot orthotics. This is primarily due to the fact that the engineering analysis
is complex and the dynamics of human gait characteristics are difficult to model. This
work further extends the use of FEA to model and analyze prescription CFOs.
11
Methods
1) Nonlinear Material Property Estimation
An initial challenge when creating a FEA of CFOs results from the limited
availability of material properties. This challenge is compounded by the highly complex
and nonlinear traits of these materials. For this study, we considered the most popular
material used in typical semi-rigid style CFOs: polypropylene. Ten sample sheets of
polypropylene with varying thicknesses (2mm, 3mm, and 4mm) provided by Kintec
Footlabs were examined using ASTM standard D575-91, and D412. Material properties
were approximated through uniaxial tensile testing and resulting stress-strain behaviors
were constructed.
To capture the nonlinear material properties, the Mooney-Rivlin strain energy
function was adopted and the constants were calculated (Finney, 1988) and can be
expressed as:
S = 2(a-a-2
) (C1+C2*a-1
); G = 2*(C1+C2) (1)
where S is stress, a is the principal stretch ratio (1+dL/L), and material constants C1 and
C2 relate to the shear modulus, G = 2240MPa. The constants C1and C2 are derived from
plotting the equation:
(S/(2*(a-a-2
))) vs. a-1
(2)
These equations will form a line, where C2 is equal to the slope of the line, and C1 is
equal to the intercept at a-1
= 1. C1 = -22623 and C2 = 23743.
12
2) Geometry Construction
The replication of CFO geometry using CAD tools is extremely time consuming
and almost impossible to accurately represent. An alternative approach involves using
laser scanning technology to generate the geometry. Laser scanning captures an object’s
geometry in a point cloud surface image, simplifying the process while creating the most
accurate result. In this project, a semi-rigid CFO was scanned using a laser digitizer and
the point cloud image was converted into a solid image using 3-Matic, a commercial
imaging tool from Materialise Inc.
3) Modeling of Constraints and Loads
Engineering analyses rely on the identification of kinematic constraints and
applied forces. The mid-stance phase of gait was modeled, as this is a good starting point.
The following boundary conditions were applied: 1) zero degrees of freedom (DOF) in
the underside of the heel region; 2) zero horizontal movement on the back, lateral and
front edges; and 3) zero movement in the vertical direction on the bottom area. The arch
area was left unconstrained. Finally, a uniform surface pressure load was applied, to
simulate mid-stance, using a static large deflection model to the entire top surface. These
boundary conditions can be seen below in figure 4.
13
Figure 4: Boundary conditions applied to FEA model simulating the midstance phase of
gait: (a) Zero DOF in the vertical direction on the bottom area (arch left unconstrained),
(b) Zero DOF in the underside of the heel region in all directions, (c) Zero DOF in the
horizontal direction on the back, lateral and front edge, (d) uniform surface pressure load
applied to the entire top surface.
4) Results
The ANSYS finite element package was used to run the analyses on three
separate weight classifications (Trinidad, 2008) and the maximum von Mises stress and
deflection results were acquired (Figures 5 and 6).
Figure 5 is a plot of the applied load versus the deflection of the arch area using
three weight classifications. Each line represents a different model which only varies by
thickness. The larger the load applied, the more deflection is seen in the arch. The
thicker the model the less deflection there is.
14
Figure 5: Maximum deflection vs. the applied load by orthotic thickness.
Figure 6 shows a representative plot of the von Mises stress distribution which
shows that the maximum stress areas are around the edge of the arch.
Figure 6: Stress distribution view (a) top and (b) bottom.
5) Discussion & Summary
This research details the successful development of a simulation-based design
procedure for the systematic design of custom foot orthotics. Properly employed, these
models may enable optimal CFO designs based on a patient’s body weight, activities,
loading conditions, foot deformities, etc. Finally, these models and the corresponding
results promise to form the basis of a new generation of custom foot orthotics. The
obvious next step is to figure out if the model is estimating the CFO behavior accurately.
This leads us to the current studies verifying and validating the FEA model to correspond
to CFO behavior.
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
0 0.2 0.4 0.6 0.8 1 1.2
Applied Load (MPa)
De
lec
tio
n (
mm
)
2mm
3mm
4mm
5mm
15
CHAPTER 2
LITERATURE REVIEW
Introduction
The following section provides a review of the literature concerning orthotics
research using FEA and background on surrogate modeling. This section begins with a
comprehensive review of orthotics research through clinical studies and followed by a
more in-depth examination into the use of Finite Element modeling. It ends with an
introduction to surrogate modeling and how it can be used as a tool to complement the
current CFO prescription process.
Orthotics
As mentioned above, Custom Foot Orthoses (CFOs) are often used as an
acceptable method of managing injuries, and while they usually produce encouraging
outcomes, it still remains unclear how the dynamics of the lower extremity are influenced
by the device (MacLean et al., 2006). Many previous clinical studies have been
performed on the effects of CFO intervention, and many have focused specifically on the
effects during running. These studies have focused on rear foot and tibial kinematics, and
both lower extremity kinematics and kinetics (MacLean et al., 2006). Variability has
been seen in study results due to two main reasons: 1) the types of subjects used; and 2)
the design of the experiment (MacLean et al., 2006).
Many research investigations have distributed identical designs to each subject in
order to limit the confusing effects of the orthotic (Mundermann et al., 2003). It has been
argued that using the same orthotic design for all subjects could be just as or more of a
puzzling factor, given that the resulting device may not be comfortable or suitable for
16
each subject’s needs (MacLean et al., 2006). CFOs are usually prescribed by podiatrists,
physical therapists and sports medicine physicians (Root, 1994) and then manufactured
from a volumetric impression of the foot by a certified laboratory to address the specific
needs of the patient. CFO research has not always included subjects who would normally
be candidates for the intervention; many studies have utilized healthy or injury-free
subjects (MacLean et al., 2006).
The main findings from CFO clinical studies have been: significant decrease in
maximum rearfoot eversion angle (Bates et al., 1979; Smith et al., 1986; MacLean et al.,
2006), decrease in maximum rearfoot eversion velocity (Smith et al., 1986; MacLean et
al., 2006), decrease in maximum internal ankle inversion moment (Mundermann et al.,
2003; Williams et al., 2003; MacLean et al., 2006), decrease in impact peak and
maximum vertical loading rate (Mundermann et al., 2003), and decrease in maximum
tibial internal rotation angle (Nawoczenski et al., 1995).
Although these studies have shown results, these results have been considered
somewhat ambiguous due to the questionable experiment designs mentioned above.
Therefore, the actual effects orthotics have on the kinematics of human locomotion
remain unclear due to the fact that it is either extremely difficult, very time consuming or
not possible to design a study that will incorporate the appropriate subjects and
investigations. This has limited researcher’s ability to draw strong conclusions regarding
the design and effect of CFOs. This is where modeling and analysis can assist in the
progression and facilitation of this research. Modeling allows for the ability to control
subject variability, thereby minimizing some of the uncertainty found in the current
literature.
17
Orthotics & Finite Element
It has been publicized by many researchers that biomechanical factors play a
crucial role in the study of orthotics. (MacLean, 2006) Little biomechanical data is
available in the literature to assist in understanding how such factors can effectively be
applied to the development of orthoses. It is possible to simulate foot motions, change in
material properties, different loading conditions, and different orthotic conditions using
accurate FE models. These models can be altered relatively easily, making it possible to
further our understanding on the influence that the device has on biomechanical factors.
Currently, the majority of FEAs on orthotics have focused on two types of
orthotic inserts: Ankle-Foot Orthotics, or AFOs, and accommodative orthotics (Figure 7).
AFO research has focused on analyzing stress points found in the device when in use.
This research has allowed for optimal designs leading to the reduction of orthotic fracture
and increase in patient compliance. On the other hand, research on accommodative
orthotics has primarily focused on the reduction of peak plantar pressure in the hopes of
preventing foot ulcerations. Both will be addressed further in the following sections.
18
Figure 7: (a) ankle-foot orthtotic, (b) accomodative orthotic
Ankle Foot Orthotic Finite Element Analysis Research
AFOs are designed to help control the motion of the ankle while offering support
to the foot. They are often used to treat conditions such as drop-foot, posterior tibial
tendon dysfunction, severe flatfoot, arthritis of the ankle and/or foot, ankle sprains, lateral
ankle instability and tendonitis.
There are three major objectives for the design of an AFO. The first is to control
motion, correct deformity, and compensate for weakness, thereby restoring normal
function and ability. The second objective is to make the orthotic as comfortable to wear
as possible in order to increase patient compliance. The third objective is to minimize the
abnormal appearance of the orthotic. Most advanced AFOs have been unable to improve
on all three objectives. The goal of most early research was to either reduce the weight or
bulkiness of the orthotic to increase patient compliance or strengthen the weak spots that
tend to fail due to high stresses applied by the foot.
Early studies using FEA on AFOs investigated the response of the ankle with and
without orthotics (P.C. Lam et al., 1986) by analyzing peak stresses and deformation
patterns. Later studies used FEA to predict loads at which AFOs become unstable and
analyze the stress distributions (D. Leone et al., 1991; T-M Chu et al., 1991). More
19
recent work on AFOs includes a study using FEA to suggest improvements on lowering
the weight and improving the comfort of an orthotic by evaluating real time pressure
between the subject and the orthotic during routine actions (walking, chair rise, stair
climb, pivoting) via a resistive pad. From the collected data, an accurate model of the
orthotic was created and the stress caused by the above activities was evaluated, leading
to modification suggestions to reduce orthotic weight (Khamis S. Abu-Hasaballah et al,
1997). Most recently, FEM was used to determine optimal design features (strut
thickness) prior to implementing into the AFO manufacturing process (Faustini, 2008).
Accommodative Orthotic Finite Element Analysis Research
Accommodative orthotics are primarily used for the prevention of foot ulcers
through the reduction of plantar pressure levels by redistributing the stresses between the
foot and orthotic. Foot ulcers are a serious problem for people suffering from diabetes as
it can lead to foot amputation and ultimately death. Neuropathy and vascular disease are
complications associated with diabetes, and although both may be present, the pathology
results in either sensory deficit (neuropathy) or vascular impairment (vascular disease).
Skin ulcerations are a result of chronic sensory neuropathy. A protective threshold is
when a person possesses adequate sensation to determine when his or her body is at risk
of harm from an outside source. At any point below this threshold, there is inadequate
sensation to signal the brain to potential harm. When the protective threshold is lost, this
allows repetitive, painless trauma to occur to soft tissues and skeletal structures which
may further increase the sensory deficit.
Friction, pressure and shearing are the three causes of stress and of great concern
for diabetics. Friction is the surface resistance of one body sliding over another. Blisters
20
are caused by fast and constant friction; the opposite causes calluses. The vertical ground
reaction forces applied to the foot is referred to as pressure. Ischemia can be caused by
constant pressure and can result in necrosis (tissue death). Shearing is a combination of
friction and pressure and can occur when two surfaces slide over each other, with
pressure being applied perpendicular to the direction of movement. This force is often
produced during normal gait. These forces can cause potential injury to the bones and
soft tissues (joint subluxation and skin ulceration).
Orthotic therapy is intended to decrease Ground Reaction Forces (GRF) applied
to the foot. An exact mold of the foot is extracted and if localized areas of pressure
occur, the GRF can be reduced by elevating adjacent areas, such as with metatarsal pads.
Distributing GRF over a greater time period will decrease shearing. For example, soft
materials will slow the foot by increasing the vertical distance the foot travels before
coming to rest. If the orthotic materials are rigid the poor shock absorption and non-
accommodative properties will not be helpful for these patients. Corrective components
of orthoses aim at decreasing unnecessary pressure on the foot by limiting excess motion
and maintaining an unstable foot in proper alignment.
Reduced plantar pressure levels to prevent foot ulcers can be achieved with in-
shoe orthoses. They reduce the pressure at bony prominences, especially under the
metatarsal heads. Although this method is readily used, very little actual quantitative
information is available regarding the effect of thickness and influence of soft tissue
characteristics on the cushioning effect of these orthoses. FEA has been used to analyze
accommodative orthotics mostly in the late 1990s and most recently in 2006. Nonlinear
material properties are difficult to model and only with recent computing advancements
21
has this become more common. The first study used FEA to compare insoles of varying
thicknesses by calculating peak plantar pressures and validating these models and values
through clinical measurements (Lemmon et al., 1995). Two years later this same group
investigated alterations in pressure under the second metatarsal head as a function of
insole thickness and foot tissue thickness. The group found that orthoses reduced plantar
pressure and offered techniques which allowed for a better approach to understanding
plantar cushioning as well as the principals involved in the design of therapeutic footwear
(Lemmon et al., 1997). More recently in Chen et al.’s 2003 research, FEA was used to
study the effects of total contact insoles on plantar stress redistribution by analyzing
different stress reduction and redistribution. This research allowed for recommendations
to be made on the effectiveness of accommodative orthotics.
In 2005, Erdemir’s group took a comprehensive look at plantar pressure relief in
footwear with compliant material plugs. 36 plug designs were looked at: 3 materials, 6
geometries, and two placements. In Barahi, 2005, they looked at reducing the plantar
pressure levels specifically under the hallux. 3-D models of the insoles were constructed
and analyses were run comparing 4 different materials: silicone gel, plastozot, plyfoam,
and EVA. And finally in Actis 2006, they developed a patient-specific mathematical
model of the second and third rays of the foot. Different models of the foot were
constructed with varying levels of detail with the foot in the push off position. They used
quasi-linear material properties for the TCI by taking the slope of first and last linear
portion and a 2nd
order spline to connect the two linear portions. There have been some
recent studies on the foot insole interaction, but these studies have primarily been
emphasizing the finite element model of the foot as opposed to the orthotic.
22
Currently nothing comparable exists in the literature regarding the application of
FEA to the understanding and design of CFOs and, more specifically, semi-rigid style
custom foot orthotics.
State of the Art
In the most recent and sophisticated research, a detailed FEA model of a foot was
created and the stress distribution under the foot for several types of orthotics were
studied (Cheung, 2008) Cheung’s research focused on different combinations of
structural and material design factors on plantar pressure distribution. The sensitivity of
six design factors, arch type, insole and midsole thickness, insole and midsole stiffness,
and custom molded shape, of foot orthosis on peak plantar pressure relief. Custom
molded shape was found to be the most important factor in reducing peak plantar
pressure. The insole stiffness was found to be the second most important feature for peak
pressure reduction. Statistics based FE method was found to be an effective approach in
evaluating and optimizing the design of foot orthosis. The research developed in this
research focus on enhancing the custom molded shape and looking at varying the arch
height to analyze the effects on stiffness.
Metamodeling
Modeling is a tool used to complement the design process and make it more
efficient. Models are abstractions from reality (Hazelrigg et al., 1999). They simulate
the behavior of a physical system in specific circumstances using performance variables
in order to project the behavior of that system. These models offer a cost efficient way to
view system performance and offer practical solutions without the use of physical
models. Many engineering analysis models can be computationally costly; some may
23
take hours or days to run a single simulation (Shao, 2006). Metamodels (or Surrogate
models) are models of models that are one level of abstraction away from an original
design or analysis model and can be used to make computations more efficient.
Although they are simplifications of models, they still maintain critical characteristics of
the original models (Shao, 2007, Simpson, 2004).
Although computer-model-based design methods have been increasingly popular
and almost necessary in the engineering design process, the creation of complex and
accurate numerical models can still require exorbitant computational costs. Complex
models such as a coronary stent expansion simulation by Shao 2007 was reported to take
91 hours to complete on a Pentium 4 computer with 3Ghz CPU and 780MB available
physical memory, and 512KB cache memory, (Shao & Krishnamurty, 2006,
Krishnamurty & Shao, 2005) even if only 1/24th
portion of bare metal stent was included
in the model. Shao speculates that “if the pressuring balloon, coronary arterial tissues,
and the irregularity in human tissues were considered in the stent model, the computer
simulations would have taken weeks or even months to get the results” (Shao, 2007).
Numerical models with such complex high definition details make design optimization
incredibly difficult and in some cases impossible. In order to arrive at an optimal design
anywhere between a dozen to hundreds of simulation runs are usually required. These
challenges are therefore compounded if each simulation requires several updates at each
feasible point before successful convergence (Booker, 1999). Alternatively, surrogate
models built as statistical approximations to these computationally expensive numerical
models have been increasingly popular in accelerating the process in design and
optimization.
24
The fundamental idea in surrogate-model-based design is to develop a less costly,
more efficient mathematical approximation to the costly numerical model based on
simulation data collected from a small number of runs of the more expensive numerical
model. This surrogate model is then used alternative to the original simulation model to
assist in “design space exploration, design optimization, reliability analysis, etc…” (Shao,
2007) (Krishnamurty & Wilmes, 2004, Simpson, 2004). Surrogate model simulations
frequently take less than one second. Designers can gather extensive information about
unknown systems using surrogate models without paying large computational costs.
Once a high-fidelity numerical model is created, information about the system to
be designed is collected through a small number of sample points corresponding to
various design alternatives. A surrogate or metamodel (typically used interchangeably), a
statistical approximation of the numerical model is then created from the sample data.
Successive designs and optimizations are executed using the surrogate model. The
subsequent design modifications are more efficient than in the numerical-model-based
design approach. Therefore efficiency of design optimization and reliability analysis can
be greatly improved with a good surrogate model. Designers can use surrogate models to
get extensive insight into unknown system without paying a heavy computational price.
Summary
Traditionally, custom designing orthotics has been a process primarily using
empirical methods. Very little actual quantitative information is available regarding the
effectiveness of CFO’s and little scientific evidence is available to provide guidelines for
persons who prescribe insoles. Engineering modeling and analysis techniques allow for
the possibility of significant contributions to CFO research community as well as a
25
greater understanding of interactions between foot and orthotic. Due to the challenge
involved in the multi-layering of the soft and hard materials of the CFOs, only the
polypropylene support phase will be looked at in this research.
26
CHAPTER 3
METHODOLOGY
General Introduction
The goal of this dissertation is to introduce alternative methods of researching
CFOs; by evaluating how engineering modeling, analysis and optimization can enhance
the design and advancement of CFO’s. Very little definitive quantitative information is
available regarding the effectiveness of custom orthotics and little scientific evidence is
available to provide guidelines for persons who prescribe insoles. This research
investigated methods that will allow for an increase in scientific evidence to CFO
influence as well as provide practical applications of these methods to practitioners.
This goal was accomplished in the two separate studies outlined below.
Study 1
Introduction
This study involves the verification of FE models of CFOs with experimental
trials, the validation through clinical trials, and the investigation on the influence of
arch height and load location on the deformation of the arch.
The model has been verified through Instron® testing, and the completion of
the validation took place in the Biomechanics Lab in the Kinesiology Department.
A clinical trial with three subjects allowed us to validate the simulated behavior of
the model to accurately represent the behavior of CFOs. The final analysis was to
investigate the effect of varying desired variables in the FEA by varying arch height
and load location in the model. By varying these factors we can investigate how
27
these factors affect the performance of the orthotic, as well as show the flexibility of
the model.
Verification and Validation
The purpose of this study was to develop refined FEA simulation models to
mimic CFO behaviors under loading and validate the results through experimental data.
These tasks are important because the FEA model of the custom foot orthotic is the first
of its kind. CFOs have not yet been studied using the Finite Element Method (FEM) to
date. These models provide a starting point for further research on CFOs using FEA.
Verifying and validating these models will ensure that the model mimics real orthotic
behavior.
A Finite Element model was created using the ANSYS package where a custom
foot orthotic is modeled with specific geometry and material properties, as described in
the previous work (Chapter 1). We will be discussing the task of running an
experimental trial to verify the model.
Experimental set-up and stress analysis
We wanted to perform an experiment in a controlled setting where we could test
and verify the material property behavior of the FEA model. Therefore, the
experimental trial was performed on an Instron machine in the materials lab of the
Mechanical and Industrial Engineering department at the University of Massachusetts
Amherst. Multiple orthotics of varying thicknesses were clamped onto a metal plate on
the heel and front edge of the orthotic. Then the plate was placed securely into the
Instron machine (Figure 8).
28
Figure 8: Experimental set-up for orthotic in Instron machine in MIE materials
laboratory
Since the primary area of interest is the deflection of the arch area, the load was
concentrated so that it would deflect the arch while minimizing the effects of the
compression of the material. Therefore, a 15mm diameter point load was applied onto
the highest point of the arch (measured to be 7.5mm from the front edge) of the orthotic.
Three different loads were applied (60N, 140N and 200N). These values were chosen to
describe a range of loads, but loads heavier than 200N resulted in the buckling of the
orthotic. The model was then run in ANSYS WB to simulate the Instron experimental
tests with a concentrated 15mm area load.
Clinical trial and data collection
This model was also tested by running clinical trials and comparing them to the
model values. The model was run in ANSYS WB with a uniform load applied to the
29
top surface in order to simulate the mid-stance phase of walking and compared to the
clinical trial data (Figure 9).
Figure 9: Uniformly distributed surface load application
In order to test and validate the simulation results, a clinical study was
performed with three subjects of varying weight (Light: 0.040MPa (88lbs), medium:
0.073MPa (160lbs) and heavy 0.138MPa (330lbs)). The study involved two sessions.
During the first session, the subjects came into the lab to measure their weights and cast
their feet. The cast of their feet was sent to Kintec Inc. for the manufacturing of the
orthotics. Once the orthotics were ready, the second session involved the data
collection. Specifically, six reflective markers were placed on the rim of the orthotic
placed at the following positions: 1) 75% of the total length (medial), 2) 50% of the
total length (medial), 3) 50% of the total length (lateral), 4) center of the heel cup
(medial), 5) Center of the heel cup (lateral), 6) on the back rim at 50% of the total width
(Figure 10). The markers were placed in these areas in order to decipher the 3
dimensional motion analysis raw data. Markers 1, 2 and 4 gave the curvature of the
30
arch which allows the measurement of the change in height of the arch by looking at
marker #2 as compared to markers 1 and 4 in the z-direction. Marker number 6 gives
the midpoint of the orthotic and the motion of the heel or rear foot as well as the change
in marker #2 compared with marker #6 in the x-direction. The lateral markers (3 and
5), give a frame of reference when looking at the data from the front or back and shows
the change in arch height in the y-direction.
Figure 10: Orthotic used for clinical trials including tracking markers
With the markers placed on the orthotic, the orthotic was adhered to the
subject’s foot using double stick tape. The subjects were asked to complete three tasks:
1) stand on one leg, 2) walk and 3) run. Each of these tasks was completed with the
orthotic attached to one foot and only that foot touching the force platform.
Prior to collecting data, the orthotic was placed on the force platform in the
center of the cameras to collect a non-weight-bearing data set. The cameras recorded
the position of the reflective markers. This was done to get a baseline reading of the
31
marker positions and arch height of the orthotic. This task was repeated 2-3 times. For
the first task, the subject was asked to stand on the force platform on the one foot that
has the orthotic attached to it. The subject used a stick to assist with balance, and the
subject held this position for approximately 10 seconds. That task was repeated
approximately 5 times. For the second and third task the subject was asked to walk then
run from one end of the laboratory to the other across the force platform making sure to
place the foot with the orthotic attached to it on the force platform. This took
approximately 10 seconds and was repeated 10 times for each activity. A three
dimensional movement analysis system quantified movement patterns of the reflective
markers during each trial. Subsequently, the clinical study results were compared to the
simulation results to further refine and validate the simulation models.
Subjects
An informed consent (Appendix) was required for the subject. The individual
must be 10 - 60 years of age, healthy as per a Modified Physical Activity Readiness
Questionnaire (Appendix) and not have had any surgeries in the lower extremity or foot
area. This study consisted of the subject completing one task for the data collection
phase. This session took approximately 30 minutes.
Experimental set-up
A custom semi-rigid style orthotic was manufactured for the subject. The
orthotic was measured on its own for baseline measurements of the arch height. The
subject stood on the device and the weight was translated down to deform the foot
and in turn the arch. The amount of deflection the arch experiences was recorded.
There were tracking markers on the arch and were recorded using the camera system
32
in the biomechanics lab at the University of Massachusetts Amherst. The deflection
was compared to the values seen in the FE model.
Data reduction
Marker traces for each trial will be processed and identified in Qualysis TM
software, and computations will be performed in Visual 3DTM
and MatlabTM
software.
Based on the movements of the markers, three dimensional movement patterns for the
arch of the CFO were computed. Starting with a baseline measurement of the CFO
position without any weight applied, and the subjects’ weight was applied to the CFO.
The position of the triangulation of markers outlining the curvature of the arch are
captured in both the non-weight bearing position and the applied weight position and
compared. The change in height of the highest marker was used as the resulting value
for deflection of the arch.
Cantilever beam test
Many researchers have compared the human foot and in particular the arch area
to a beam (Cavanagh, 1987& Steindler, 1955). Originally the material properties for
this finite element model were calculated using uniaxial tensile tests (mode 1) to get the
modulus of elasticity through tension. Upon further thought and realizing that the
majority of the deformation seen on the arch of the orthotic was through bending and
shear similar to what a beam would experience, the decision was made to test the
material properties through a cantilever beam test (Mode 2).
The cantilever beam test was performed using a 1”x 3” rectangular bar of
polypropylene which was clamped into an Instron machine (Figure 9). A strain gage
was glued onto the polypropylene bar and a 5N load was applied at a distance of 56mm
33
from the clamp. The strain was collected and the stress calculated using the bending
stress equation (Juvinall, 2006):
(2)
where M is the bending moment, h is the thickness of the sample (in meters), and I
is the moment of inertia. This equation can be derived for our purposes to a more
useful format:
(3)
where P is the load applied (5N), Δl is the length of the active portion of the beam
(l3 – l2 = 0.046m), b is the width of the beam (0.02m), and h is the thickness of the
beam (0.003m). These calculations yielded a stress of σ =7.67MPa. The average
strain measured by the Instron was ε = 0.003134 after 5 trials with a Standard
deviation of .000012. Young’s modulus was calculated from the stress and strain
collected to be 2446MPa. For the uniaxial tensile testing E was calculated to be
4480MPa for the 3mm heated polypropylene which is about double the modulus of
elasticity calculated using the cantilever beam test.
Figure 11: Cantilever beam material set-up including 1”x 3” rectangular bar of
polypropylene. A strain gage was glued onto the polypropylene bar and a 5N load was
applied at a distance of 56mm from the clamp
34
Varying load location and arch height
In order to alter the arch height and concentrated load location the model was
transferred into Solidworks. A Solidworks VBA macro developed by Krishna
Samavedam was used to generate and parameterize the control points for splines. The
arch height ranged from 16mm (from the floor/base fixture) to 25mm (from the
floor/base fixture) as seen in Figure 12 (a). The concentrated load location ranged from
55mm (from front edge) to 80mm (from front edge) as seen in Figure 12 (b).
(a)
(b)
Figure 12: ANSYS WB model, (a) Medial side view showing arch alteration lines, (b)
top view showing Instron 15mm diameter load location and alteration lines as well as
30mm diameter constraints.
35
The following operations are performed when this macro runs.
1. The control points that are sampled on the STL model of CFO are imported into
Solidworks.
2. Three lines are drawn along X, Y and Z axes to control the coordinate location
of the control point.
3. Dimensions are assigned for each line.
4. Parametric equations are generated between the dimensions and the parameters.
This Solidworks model was attached to the ANSYS WB model and the modifications
were made automatically depending on the desired design point to be analyzed.
Summary
This study evaluated the use of FEAs ability to investigate design changes and
their effects on the device by investigating the influence of the arch height in reducing
stresses. This study also validates the FE model previously described in the
introduction by evaluating the effectiveness of these models to mimic CFO behavior.
The change in arch height (deflection) during the clinical trials is compared to the
change in arch height (deflection) in the model.
Study 2
Introduction
The purpose of metamodels is to provide approximation models for structuring a
design problem, increase the speed of computing, and make the design process more
efficient. They simulate the behavior of the actual system as close as possible in a more
economical way by using simpler approximations through least squares regression or
interpolation methods. In this study metamodels of the FE model were developed and
36
a tool for clinicians’ to replace empirical tables (Table 2) to use in the office to
compliment current prescription processes was suggested.
Experimental set-up
The objective of a routine statistics-based performance model is to mimic the
systems input/output behavior based on the collected data distributed over the design
space. ANSYS has the ability to run surrogate models using four different methods:
Full second order polynomial standard Response Surface Method (RSM), Kriging, Non
Parametric Regression (NPR), and Neural Networking (NN). This study compares the
accuracy and efficiency of each of the four methods.
Response surface uses a least square regression fitting method, whereas Kriging
uses the same method to fit the overall trend of the data, while also calculating the
errors associated with the correlation of the distance between data points. The non
parametric regression “is a form of regression analysis in which the predictor does not
take a predetermined form but is constructed according to information derived from the
Table 1: Sample weight to thickness ratio guideline for orthotics prescribers
37
data.” (McCune, 2006) The Neural Networking method is inspired by the structure
and/or functional aspects of the human brain. It has the ability to learn and can change
its structure based on pattern adapted from the learning phase. Neural Networks are
used to model relationships between inputs and outputs or to find patterns in data. The
Neural Networks learning capabilities work by using a set of observations to define a
relationship between the input and output data in some optimal sense (Smith, 1993).
ANSYS Workbench explores a relationship between the design variables X1
(arch ht) and X2 (orthotic thickness) and the response variable Y (deflection) to form an
approximation model (a response surface) of the data. Three design surfaces were
created for each loading scenario; each scenario was used to create a design surface for
the three different load classifications designated as: light (0.040MPa), Medium
(0.074MPa) and Heavy (0.138MPa). The design surface was created using 15 design
points, the design points were chosen at the three arch heights: 16 (low), 20.5 (medium)
25 (high), for each orthotic thickness: 1.5mm, 2mm, 3mm, 4mm and 5mm. These
surrogate models can be used to explore other uses and effects of custom foot orthotics
without running the full finite element model. Additional variables may also be added
to explore other situations without changing the finite element model. Although these
surrogate models are proven to be more efficient by saving time, the loss of accuracy
during the “approximation” was investigated with respect to the usability of the results.
Validation
Once the surrogate models were built, each model was validated and the results
compared to its deviation from the original finite element model. The validation of the
models was executed by building another data set using a rotatable central composite
38
design (Montgomery, 1997). This new data set contains only input values and the
output values were from the surrogate model. These output values were compared to
the original finite element model output values and compared by % difference. It is
important to know how much accuracy the surrogate model is maintaining while
approximating the full model. The rotating central composite design will allow the
retrieval of data from each quadrant of the approximation surfaces directly at points
used to create the model as well as at points not used to create the model.
Model Building:
The models were built in ANSYS WB using four different methods; Response
surface, Kriging, non parametric regression and neural networking. The different
methods use different approaches to creating a relationship between the input and
output data used to build the model. Three different arch heights and five different
orthotic thicknesses were used to build the models. The data used are shown in the
table below:
39
Table 2: Data used to build the surrogate models for each scenario
Light (0.040MPa)
Medium (0.074MPa)
Heavy (0.138MPa)
Arch Ht Thickness deflection
Arch Ht Thickness deflection
Arch Ht Thickness deflection
16 1.5 6.79
16 1.5 12.35
16 1.5 23.15
16 1.5 6.79
16 1.5 12.35
16 1.5 23.15
16 2 3.92
16 2 7.13
16 2 13.36
16 3 1.85
16 3 3.37
16 3 6.31
16 4 1.12
16 4 2.04
16 4 3.83
16 5 0.79
16 5 1.44
16 5 2.71
20.5 1.5 8.57
20.5 1.5 15.59
20.5 1.5 29.24
20.5 2 4.49
20.5 2 8.17
20.5 2 15.32
20.5 3 1.97
20.5 3 3.59
20.5 3 6.73
20.5 4 1.20
20.5 4 2.18
20.5 4 4.09
20.5 5 0.86
20.5 5 1.57
20.5 5 2.94
25 1.5 10.46
25 1.5 19.03
25 1.5 35.69
25 2 5.39
25 2 9.81
25 2 18.40
25 3 2.26
25 3 4.11
25 3 7.71
25 4 1.34
25 4 2.45
25 4 4.58
25 5 0.96
25 5 1.75
25 5 3.28
Summary
There is much that is still unknown in the general area of CFOs. Modeling is a
tool that can greatly improve the methods in which are used to determine the
effectiveness of orthotics. These surrogate models will not only provide tools to make
the CFO research and prescription process more scientific and optimal, but it may also
provide a possible tool for clinicians to use in the office to replace the empirical tables
they currently use. This study determines the cost-to-benefit ratio of the four methods
used to build each surrogate model by comparing the efficiency and accuracy of each
method. While these methods have never been used in the study of CFO, these studies
will allow for the application of much more efficient design and development of CFOs.
40
Summary
It is important to understand which methods are the best methods to proceed
with in future research. Although clinical trials are currently the most common
technique used for CFO investigation, previous work in Trinidad (Trinidad, 2008)
showed that FEA models may be less time consuming and allow for more variability in
the design of experiments and provide concrete results to the current data collected
through clinical trials. The incorporation of surrogate models may allow for even more
freedom and efficiency and allow for better practical application.
41
CHAPTER 4
RESULTS
Study I
Instron Verification
The Instron tests were run using different size orthotics with various thicknesses,
applying a 15 mm round point load to the highest point of the arch (approximately 50%
of the length on the medial side of the orthotic). This point load was also applied to the
FEA model and the difference in means, standard deviation, and sample size were
compared using a paired-t test. The resulting P-value was used to measure the precision
in the sample mean, since we are estimating the overall population average for our
comparisons. The P-value is defined as the probability that one data set is as much as or
more extreme than the other observed data set if the null hypothesis was true (Evans,
2011). The null hypothesis states that the FEA model data and the clinical trial data
are equivalent:
H0: Dfea=Dclin,
Ha: Dfea≠ Dclin
An alpha level of 0.05 was used as is conventional in clinical studies. “Statisticians
have found that approximately 95% of all randomly selected sample means fall within 2
units of standard error from the population mean” (Batavia, 2001, Ch. 11).
There were three different size orthotics used for these studies; a small
(women’s size 5), a medium (women’s size 7.5) and a large (men’s size 16). Each size
orthotic was made in 3 different size thicknesses (between 2mm, 3mm, 4mm and 5mm).
Corresponding models were created in ANSYS and the arch deflection was compared.
42
The means, standard deviation and percent error are compared between the
experimental (Instron) tests and the ANSYS model trials. The loads applied were 60,
140 and 200 N applied as a 15mm concentrated load on the highest point of the arch.
The results between the ANSYS model (Trinidad, 2008 & Chapter 1) and the
Instron tests show very good agreement and verify that the ANSYS models can
reproduce real world behavior as can be seen in Table 4 below. The percent differences
between the two sets of results are mostly less than 5%, and the probability values
indicated that there is no significant difference between the two methods. There are
three values that are above 5% difference, (Medium orthotic, low load, and large
orthotic, low and medium load) but they are still below 15% difference and the p-values
still indicate no significant difference in the values.
Table 3: Deflection for Instron experimental tests vs. ANSYS WB deflection values.
Small
Load (N)
Instron Deflection δ (mm)
Standard deviation
Standard error
Model Deflection δ (mm)
% Difference P-Value
60 1.532 0.04 0.02 1.56 1.81 0.26
140 3.222 0.11 0.05 3.24 0.56 0.77
200 4.376 0.21 0.09 4.24 3.16 0.29
Medium
Load (N)
Instron Deflection δ (mm)
Standard deviation
Standard error
Model Deflection δ (mm)
% Difference P-value
60 2.392 0.80 0.36 2.13 11.59 0.56
140 5.078 0.28 0.13 4.97 2.15 0.50
200 6.97 0.41 0.18 7.1 1.85 0.57
Large
Load (N)
Instron Deflection δ (mm)
Standard deviation
Standard error
Model Deflection δ (mm)
% Difference P-value
60 2.64 0.88 0.40 2.34 12.24 0.54
140 5.544 1.41 0.63 6.44 14.95 0.29
200 7.814 0.53 0.24 7.45 4.83 0.26
43
Clinical Validation
Uniform Surface load
The clinical validation was run using three subjects light, medium, and heavy;
each subject had different orthotics of varying thicknesses to test. The light subject had
a 2mm, 3mm, and a 4mm thick orthotic. The medium subject had a 2mm, 3mm, and a
4mm thick orthotic to use. The Heavy subject had a 3mm, 4mm and a 5mm thick
orthotic to use. Although we collected data for all 9 scenarios, we chose to focus on the
3mm and 4mm data in order to compare all three subjects to each other for each
condition. Their body weights were converted into a pressure load (Pa) and those
values were inserted into corresponding models that were created in ANSYS as a
uniform surface load and the arch deflection was compared to the clinical trials. We
began with a simplified uniform surface load, although a person’s weight is not
uniformly distributed throughout the bottom of their foot, it was a reasonable starting
point (Figure 7). Statistical data were calculated (% difference, mean, standard
deviation, and P-value) between the model and clinical data.
Although, the clinical trials show less agreement with the ANSYS model as
compared to the experimental Instron trials, the agreement is satisfactory as the majority
of the P-values are greater than 0.05 demonstrating that there is no significant difference
between the two data sets except for the heavy 3mm condition.
44
Table 4: Comparison of clinical trials data vs. ANSYS WB uniform load deflection
values.
3mm (2.78mm)
Load (Pa)
Clinical deflection CL_SD
Model deflection
% Difference in Means P-value
δ (mm) δ (mm) α = 0.05
Light 2.22 0.52 2.10 5.56 0.68
Medium 3.00 0.25 3.84 24.56 0.13
Heavy 5.56 0.06 7.25 26.37 0.02
4mm (3.78mm)
Load (Pa)
Clinical deflection CL_SD
Model deflection
% Difference in Means P-Value
δ (mm) δ (mm) α = .05
Light 1.06 0.18 1.20 12.39 0.47
Medium 1.82 0.39 2.20 18.67 0.15
Heavy 3.95 0.34 4.20 6.13 0.33
Varying distribution load
After running the model with a uniform surface load it was observed that the
agreement was satisfactory but could be improved if the loading characteristics in the
model were adjusted. The uniform surface loading was a simplified loading
distribution, but an appropriate place to start from. While investigating the loading
characteristics and weight distribution of a normal person during the stance phase of
walking (Cavanagh, 1987 and Cheung, 2005), it was noted that the weight is not
uniformly distributed but distributed at different levels throughout the foot, generally
separated by heel, forefoot and midfoot, in that order. A much smaller percentage of
the weight is seen in the arch of a foot (midfoot area). Since some of the research has
reported that normal stance is generally observed as approximately 25% – 45%
(Morton, 1953) of the weight in the heel, and some has reported that 60% of the weight
is loaded in the heel, we decided to redistribute the weight looking at both theories
45
(Figure 13). The first redistribution was accomplished by applying 30% of the weight
in the heel section of the model and the second by applying 60% of the load in the heel.
The orthotic is only a three-quarter orthotic leaving a portion of the forefoot
unrepresented. The weight was redistributed first by dividing 30% of the weight in the
back of the orthotic and the other 70% of the weight in the rest of the model, and second
by dividing 60% in the back of the orthotic and 40% of the weight in the rest of the
model.
Figure 13: Split load application (30% of load applied to back/heel portion of the
model)
Making this adjustment showed much improvement in the agreement with the
clinical trials showing more consistency and reducing the error between them. These
distributed load models seem to be a slightly more accurate representation of normal
human loading behavior and the accuracy between the clinical trials and the model
increased for most of the scenarios. The accuracy increased by decreasing the percent
difference while still maintaining a p-value greater than 0.05 for most of the scenarios.
Tables 6 and 7 show the comparison of the clinical trial data to the ANSYS 30/70 load
46
distribution and 60/40 load distribution respectively. A further, more comprehensive
investigation of the loading distribution is required to further improve upon the loading
characteristics in the model, but this is an appropriate next step.
Table 5: Comparison of clinical trials data vs. ANSYS WB varying distributed load
deflection values 30/70. Light = 0.040MPa (88lbs), Medium = 0.074MPa (160lbs),
Heavy = 0.138MPa (300lbs)
3mm (2.78mm)
Load (Pa)
Clinical deflection CL_SD
Model deflection
% Difference in Means P-value
δ (mm) δ (mm) α = .05
Light 2.22 0.52 1.59 33.24 0.09
Medium 3.00 0.25 2.89 3.81 0.65
Heavy 5.56 0.06 5.41 2.68 0.18 4mm (3.78mm)
Load (Pa)
Clinical deflection CL_SD
Model deflection
% Difference in Means P-value
δ (mm) δ (mm) α = .05
Light 1.06 0.18 0.94 11.83 0.52
Medium 1.82 0.39 1.71 6.29 0.61
Heavy 3.95 0.34 3.21 20.62 0.06
47
Table 6: Comparison of clinical trials data vs. ANSYS WB varying distributed load
deflection values 60/40. Light = 0.040MPa (88lbs), Medium = 0.074MPa (160lbs),
Heavy = 0.138MPa (300lbs) 3mm (2.78mm)
Load
(Pa) Clinical
deflection CL_SD Model
deflection % Difference
in Means P-value
δ (mm) δ (mm) α = .05
Light 2.22 0.52 1.59 33.07 0.09
Medium 3.00 0.25 2.90 3.39 0.67
Heavy 5.56 0.06 5.55 0.20 0.85 4mm (3.78mm)
Load
(Pa) Clinical
deflection CL_SD Model
deflection % Difference
in Means P-value
δ (mm) δ (mm) α = .05
Light 1.06 0.18 0.75 34.25 0.25
Medium 1.82 0.39 1.37 28.44 0.10
Heavy 3.95 0.34 2.61 40.85 0.02
Study II
The four surrogate models are built for each finite element model. The following
graphs (Figures 14, 15 and 16) show the resulting surrogate models from each of these
methods over the design space for the light, medium, and heavy weight scenarios. In
the figures 14, 15 and 16 (A) is the result of the response surface method, (B) for the
Kriging Method, (C) is the non parametric regression method and (D) the neural
network method. Each surrogate model took about one second to run.
48
(B)
AA
(A)
AA
49
Figure 14: Surrogate models for the light load scenario (A) Response Surface, (B)
Kriging, (C) Non Parametric Regression, (D) Neural networking.
(D)
AA
(C)
AA
50
(B)
AA
(A)
AA
51
Figure 15: Surrogate models for the medium load scenario (A) Response Surface, (B)
Kriging, (C) Non Parametric Regression, (D) Neural networking.
(D)
AA
(C)
AA
52
(B)
AA
(A)
AA
53
Figure 16: Surrogate models for the heavy load scenario (A) Response Surface, (B)
Kriging, (C) Non Parametric Regression, (D) Neural networking.
(C)
AA
(D)
AA
54
Model Validation:
Once the surrogate model surfaces were built, a rotatable central composite
design was used to validate these models. Experimenters typically use experiment
designs that consist of trial runs at the lower and upper extreme points (Montgomery,
1997). The Design of Experiments method was chosen based on the desired regions to
be analyzed in the most efficient way. The central composite design (CCD) is the most
commonly used design of experiments methods for three main reasons. (1) CCDs can
be run sequentially. It can be partitioned into two subsets of points; the first estimates
linear and two-factor interactions, and the second estimates curvature. (2) They are
very efficient, providing a lot of information variable relationships and experimental
error in a minimum number of required runs. And (3) CCDs are very flexible. The
variability in available CCDs enables their use under different experimental regions of
interest and operability (Verseput, 2000).
There are three main types of CCDs used: face-centered (FCCCD), rotatable
(RCCD) and inscribed (ICCD). The FCCCD encompasses the extreme points as well as
the midpoints of the region. The design consists of a center point, four factorial points
and four axial (extreme) points. The dots in Figure 17 define the variables that
constitute the nine design points (experiment runs).
55
Figure 17: Two-Variable Face-Centered CCD
“The radius, designated α, determines the geometry of the design region. An α of 1.0
defines a square design geometry (a cube for three variables, a hypercube for four or
more variables, etc.).” (Verseput, 2000).
If the precision of the estimated response surface at some point x depends only
on the distance from x to the origin, not on the direction, then the design is said to be
rotatable (Oehlert 2000). The variance remains the same when the rotatable design is
rotated about the center. The rotatable CCD is shown in Figure 18.
56
Figure 18: Two-Variable Rotatable CCD
The inscribed option is a convenient way to generate a rotatable CCD that
allows the experimenter to study the entire range of the variables while excluding
unacceptable conditions at one or more of the extremes of the design region (Verseput,
2000). The downside to this method is that it restricts the actual design region. This
design is generated by locating the axial points at the lower and upper bounds of the
variables, the factorial points are then set at a specific distance from the center point
maintaining the proportional distance between the factorial and axial points and
inscribes these points into the interior of the design space (verseput, 2000) This can be
seen in Figure 19 below where the excluded portion of the region is shown in gray, as
are the excluded face-centered CCD points.
57
Figure 19: Two-Variable Inscribed CCD
The reason for using a rotatable design in this research is that it provides equal
precision of estimation of the surface in all directions. The rotatable CCD produced 9
points (as seen in Figure 18) and the output at the four surrogate models were compared
to the output for the original model at each point. Five of the nine points were used to
create each surrogate model and the other four are intermediate points spaced at
different zones of the surface. Table 7 shows the summary of results for the light
weight scenario. This table compares the differences in deflection values from the
original FEA results when using the different surrogate models. Tables 8 and 9 show
the summary of results for the corresponding medium and heavy weight scenarios.
58
Table 7: Light scenario surrogate model compared to finite element model values and
percent difference for the Response Surface, Kriging, Non Parametric Regression, and
Neural Networking methods.
(mm) (mm (mm) (mm) (mm) (mm) (mm) deflection vs. RSM
deflection vs. Kriging
deflection vs. NPR
deflection vs. NN
Arch Ht
Thickness
FEA model RSM Kriging NPR
Neural Network
% difference
% difference
% difference
% difference
16.00 3.00 1.85 1.79 1.85 1.85 1.86 3.30 0.00 0.00 0.66
18.25 2.00 4.14 4.21 4.17 3.96 4.17 1.53 0.64 4.52 0.63
20.50 1.50 8.57 8.49 8.57 8.57 8.56 0.87 0.00 0.00 0.04
22.75 2.00 4.94 4.93 4.90 4.96 4.95 0.22 0.83 0.44 0.20
25.00 3.00 2.26 2.25 2.26 2.26 2.07 0.44 0.00 0.00 8.92
22.75 4.00 1.26 1.29 1.26 1.09 1.24 2.12 0.13 14.63 1.60
20.50 5.00 0.86 0.87 0.86 0.86 0.87 0.72 0.01 0.00 0.61
18.25 4.00 1.17 1.19 1.16 0.94 1.17 1.44 0.60 21.95 0.09
20.50 3.00 1.97 2.02 1.97 1.97 1.95 2.38 0.01 0.00 0.97
Average 1.45 0.25 4.61 1.52
Table 8: Medium scenario surrogate model compared to finite element model values
and percent difference for the Response Surface, Kriging, Non Parametric Regression,
and Neural networking methods.
(mm) (mm) (mm) (mm) (mm) (mm) (mm) deflection vs. RSM
deflection vs. Kriging
deflection vs. NPR
Deflection vs. NN
Arch Ht
Thickness
FEA model RSM Kriging NPR
Neural Network
% difference
% difference
% difference
% difference
16.00 3.00 3.37 3.29 3.37 3.37 3.39 2.19 0.00 0.00 0.65
18.25 2.00 7.54 6.20 7.59 7.20 7.59 19.42 0.67 4.52 0.65
20.50 1.50 15.59 15.43 15.59 15.59 15.59 1.04 0.00 0.00 0.04
22.75 2.00 8.98 9.29 8.91 9.02 9.01 3.36 0.82 0.45 0.25
25.00 3.00 4.11 4.12 4.11 4.11 3.76 0.07 0.00 0.00 8.97
22.75 4.00 2.30 2.40 2.29 1.98 2.26 4.34 0.05 14.75 1.50
20.50 5.00 1.57 1.54 1.57 1.57 1.58 1.74 0.00 0.00 0.76
18.75 4.00 2.12 1.86 2.11 1.83 2.14 13.33 0.54 14.89 0.68
20.50 3.00 3.59 3.62 3.59 3.59 3.55 0.75 0.03 0.00 1.01
Average 5.14 0.23 3.85 1.61
59
Table 9: Heavy scenario surrogate model compared to finite element model values and
percent difference for the Response Surface, Kriging, Non Parametric Regression, and
Neural networking methods.
(mm) (mm) (mm) (mm) (mm) (mm) (mm) deflection vs. RSM
deflection vs. Kriging
Deflection vs. NPR
Deflection vs. NN
Arch Ht
Thickness
FEA Model RSM Kriging NPR
Neural Network
% difference
% difference
% difference
% difference
16.00 3.00 6.31 6.16 6.31 6.31 6.35 2.49 0.00 60.86 0.63
18.25 2.00 14.13 11.51 14.23 13.51 14.22 20.41 0.67 64.93 0.65
20.50 1.50 29.24 28.80 29.24 29.24 29.22 1.49 0.00 0.00 0.04
22.75 2.00 16.84 17.46 16.70 16.92 16.89 3.62 0.82 0.45 0.25
25.00 3.00 7.71 7.74 7.71 7.71 7.06 0.36 0.00 0.01 8.89
22.75 4.00 4.30 4.49 4.30 3.71 4.24 4.23 0.01 14.69 1.45
20.50 5.00 2.94 2.90 2.94 2.94 2.96 1.29 0.00 0.01 0.73
18.25 4.00 3.98 3.32 3.93 3.20 3.98 18.13 1.24 21.81 0.00
20.50 3.00 6.73 6.79 6.73 6.73 6.66 0.84 0.00 0.00 0.98
Average 5.87 0.30 18.09 1.51
For the light weight data set, it appears that Kriging offers the best model with
least error over the actual FEA model, followed by the response surface model, then
neural networking and finally the non parametric design. For the Medium data, the
Kriging offers the best results, followed by the neural network method, the non
parametric regression and finally the response surface model. For the Heavy data,
Kriging again has the best results, followed by the neural network, then the response
surface, and the non parametric regression model. From the results, Kriging is the only
method that is consistently accurate for all weight classes, with less than 0.3%
difference. Each of these models within each weight category also took less than one
second to run, which is an almost 200% saving of time.
60
CHAPTER 5
DISCUSSION & CONCLUSIONS
Study I
Two studies were performed, one to verify and one to validate the finite element
model; the first study involved comparing the FEA model to an Instron test. This
deterministic test is a direct proportion of load and stresses. There was no minimal
variance between the analysis and the experimental trials. The second study involved
the comparison of clinical trials to the FEA model using a uniform load applied to the
entire surface. Although the clinical trial data were limited, the results compared
reasonably well. Most of the values were within 20% difference of each other and the
majority of the P-values were indicative that there were no significant differences
between the two data sets (there were a few outliers). This loading distribution was a
starting point since in reality people do not stand with their weight evenly distributed
across the entire foot. The majority of the weight is distributed between the heel and
ball of the foot and then a smaller portion of the weight falls in the middle of the foot
(arch area) (Cavanagh, 1987, Hills, 2001, Birtane, 2004, Morton, 1953). To make the
loading conditions more accurate, the models were refined by altering the load applied
to the surface of the orthotic, making it more realistic to the actual load distribution seen
under a foot during the stance phase of walking. We split up the load by 30% of the
total weight in the heel of the foot as some research has suggested that the heel and
forefoot are more equally loaded (Morton, 1953). Since the orthotic is a three-quarter
orthotic and the model is only split into two, back end (heel) and front end (midfoot and
part of the forefoot) we split the load into two parts, 30% in the backend, and 70% in
61
the front end. This proved to yield more accurate results which reduced the difference
between model and clinical data significantly for four of the six scenarios. Specifically,
in the 3mm thick orthotic, the differences were lowered for the medium and heavy
subjects, but increased for the light subject. For the 4mm thick orthotic the percent
difference was lowered for the medium and light subjects, but increased for the heavy
subject. For this redistribution, the P-values indicated that there was no significant
difference between the model and clinical values. Although this load distribution
minimized the errors between the model and clinical data for the majority of the
scenarios, this improvement was not consistent.
A second re-distribution of the weight was applied to the model based on
Cavanagh’s findings, which reported that they found that 60% of the weight is held in
the heel. We re-distributed the load using 60% of the weight in the heel, and 40% in the
front of the orthotic. This new distribution of load proved to also decrease the percent
difference for the medium and heavy subject for the 3mm thick orthotic, but less
accurate with the 4mm orthotics, where the percent differences actually increased for all
three subjects. Although the P-value indicated that there were still no significant
difference between the model and clinical values for the light and medium loads,
concluding that this distribution of weight can still be used to model the behavior of
human loading conditions on orthotics for light and medium subjects. The heavy
scenario P-value indicated that there is a significant difference between the model and
clinical values for the 4mm orthotic.
These validated FEA models allow for the further investigation of orthotic
intervention. These models enable easy alteration of geometry and loading conditions
62
for use in practical applications. The challenges involved in these tasks included
designing experimental studies to match FEA model assumptions and criteria. When
designing the experimental and clinical trials it was difficult to apply all of the
assumptions made into the development of the model. For instance, the model does not
contain a floor suggesting there is no friction involved. Although in the experimental
setting the friction between the floor (or base of fixture) and orthotic is very small and
assumed to be negligible, there is still a difference. Another variation is the shape of the
orthotic. The orthotic in the model was built from a sample orthotic, whereas the
orthotics used in the clinical trial were specifically made for each subject. The
difference in geometry can affect the performance such as: length of the device, heel
depth, and arch height to name a few variables. The length of the device is also a
factor, as it is only 75% of the foot’s total length and since we are applying 100% of the
subject’s weight to the orthotic, the elimination of the weight held in the 25% of the
foot that is not supported by the orthotic is not taken into account. Although this is a
very small percent of the total weight, it could account for some of the discrepancies.
Finally, in the clinical trials, the subjects were holding on to a broomstick to assist with
balance. This broomstick was used to stabilize the subject in order to minimize the
shifting of weight within the foot. Although the subjects were instructed to put as little
weight as possible on the broom stick, this will still take on some of the weight and
therefore make the weight applied to the orthotic less than in the model. While there
were some challenges and differences, the final values in the model correlated very well
with the experimental and clinical tests.
63
In the experimental tests the values matched the model to less than 15%
difference with 6 out of the 9 trials below 5% difference. For the clinical trials these
values are slightly less accurate in their match up as human subjects introduce many
other factors that affect these comparisons. These values are within 5 to 26%
difference, with most values below 15% difference. The data also indicate that there is
no significant difference according to the p-values above 0.05 except for the 3 mm
heavy category which has a p-value of 0.02. The large discrepancy in the 3 mm heavy
category is due to their only being one data set for the clinical trial, which caused the
standard deviation to be very small (0.05) so any difference in values is going to seem
like a large difference when discussing in terms of standard error and p-values. In this
uniform loaded model many of the difference in values may also be attributed to how
the person stands as opposed to how the weight was loaded into the model. If they
stand pronated, supinated or put more weight in their heel it would change the value in
which the arch is deflected.
We ran a second and third set by redistributing the weight in the model which
yielded more accurate loading conditions. The comparison of values to the clinical
trials also improved with both adjustments in weight distribution. Most of these data
lowered the comparison with the clinical trial to within 20% difference with most below
12%. Most of the P-values were also indicating that the data are not significantly
different except for the “Heavy” category. In addition to the differences in loading
characteristics of obese individuals to non-obese subjects, this smaller P-value in the
heavier load category may be due to the smaller variation in the heavier loads clinical
data causing each difference to have a greater effect.
64
One of the findings was that most of the load distributions were not very
accurate with the “Heavy” category. Although none of the load distributions were
consistently accurate, the uniform load distribution seems to be the most accurate for
the heavy subjects. Table 10 below includes the complete data set including the 5mm
data to compare all three cases for the heavy subject. Another possible difference
found for heavy subjects could be that these material properties may not be suitable for
heavy subject. The material properties were assigned based on an assumption that the
materials stay within the elastic range. The heavy subjects may possibly be heavy
enough to change the properties of the materials and therefore these material properties
may not be suitable for this population, especially for the thinner orthotics (3mm and
4mm thick).
Table 10: Clinical, uniform load and distributed load deflection values compared using
standard error and P-values
Clinical
Uniform load
FEA
Distributed
load(30/70)
Distributed load
(60/40)
δ (mm)
standard deviation
standard error ±2S.E.
δ
(mm)
P-
value
δ
(mm)
P-
value
δ
(mm)
P-
value
3mm
Light 2.22 0.52 0.23 2.68 - 1.76 2.1 0.68 1.59 0.09 1.59 0.09
Medium 3 0.25 0.2 3.39 - 2.61 3.84 0.13 2.89 0.65 2.9 0.67
Heavy 5.56 0.06 0.06 5.69 - 5.43 7.25 0.02 5.41 0.18 5.55 0.85
4mm
Light 1.06 0.18 0.11 1.27 - 0.85 1.2 0.47 0.94 0.52 0.75 0.25
Medium 1.82 0.39 0.18 2.18 - 1.47 2.2 0.15 1.71 0.61 1.37 0.1
Heavy 3.95 0.34 0.17 4.29 - 3.61 4.2 0.33 3.21 0.06 2.61 0.02
5mm
Heavy 2.89 0.1 0.05 2.99 - 2.79 3.13 0.05 2.3 0.01 1.5 0
These findings seem to be consistent with the research which states that obese
individuals demonstrate significantly different plantar pressures during both standing
65
and walking (Hills, 2001) flat footedness is also prevalent in obese subjects (Hills,
2001). The research has also found that the force distribution may shift forward into the
forefoot as bodyweight increases due to adipose tissue (Birtane, 2004). Therefore, it is
believed that a uniformly distributed load is more accurate for obese subjects. “Subjects
with higher arch index (representative of a flatter foot) demonstrate greater medial arch
lowering during the midstance phase of gait, and the movement pattern manifests as
increased loading under the midfoot.” (Menz, 2006)
Conclusion
The model verification and validation has been completed successfully. The
Instron tests deflection values were less than 15% different with six of the nine values
within 5% difference with the FEA model deflection values. The P-values greater than
alpha = 0.05 also indicate that there is no significant difference between the two data
sets. Most of the clinical trial values are below 26% difference, with 4 of the 7 values
below 15% of the FEA model deflection values and improving to below 22% when the
loading characteristics were altered to be applied more realistically. The P-values
between these two data sets also indicate that there is no significant difference between
the two data sets, except for in the heavy category. The differences that arise between
the experimental, clinical trials and the model are most likely caused by assumptions
made in the boundary conditions and other factors that cannot be prevented such as the
geometry, loading conditions, material thickness, as well as the lack of friction in the
FEA model.
This research shows that engineering analysis models using FEA can be used to
mimic the results found in clinical and experimental studies. The material behavior can
66
be modeled accurately using FEA techniques. We have identified an alternate means to
estimate the critical parameters that are crucial to custom design of orthotics based on
individual’s needs and characteristics. These validation studies prove that these models
can therefore be used to gain further knowledge and insight into the effectiveness of
orthotics. These studies will greatly enhance the current CFO research. FEA is a tool
that has not been fully taken advantage of in the investigation of CFOs and will be a
vital tool to further the body of knowledge on custom foot orthotics.
Case study: varying arch height and load location
This study started with a concentrated load along the arch area. The load
was moved along the arch (from anterior to posterior from 55mm to 80mm off the front
edge) to see what effect load location had on the arch deflection. This varying load
location model showed that as the weight is concentrated more towards the front of the
arch more deflection is seen, as well as if the load is concentrated towards the back, less
deflection is seen in the arch for most of the neutral and high arch designs. For the low
arches and the neutral arch in the medium weight category at its least towards the front
of the arch. It can be seen in Figure 20 below, where (A) is the result of the 60N load,
(B) is the 140N load and (C) is the 200N load:
67
1
1.1
1.2
1.3
1.4
1.5
1.6
40 50 60 70 80 90
De
fle
ctio
n (
mm
)
Force Application Location
Varying Load Location (60N Load)
60N @ Arch Height 16
60N @ Arch height 20.5
60N @ Arch height 25
2
2.2
2.4
2.6
2.8
3
3.2
3.4
3.6
3.8
40 50 60 70 80 90
De
fle
ctio
n (
mm
)
Force Application Location
Varying Load Location (140N Load)
140N @ Arch Height 16
140N @ Arch Height 20.5
140N @ arch height 25
(A)
(B)
68
Figure 20: Effect of load location holding thickness and load constant separated by
arch ht. Only 3mm used for example, all thicknesses are similar. (A) is the 60N load,
(B) is the 140N load and (C) is the 200N load.
This is important for clinicians to know because if someone front loads their
weight they may need a thicker orthotic, whereas if someone who back loads their
weight may be able to get away with a thinner orthotic.
Next, the orthotic was fully loaded with a uniformly distributed pressure load
matching the weight of the subjects. We then looked at the effect of the arch height on
the arch deflection. As shown in Figure 21 below:
3
3.5
4
4.5
5
5.5
40 50 60 70 80 90
De
fle
ctio
n (
mm
)
Force Application Location
Varying Load Location (200N Load)
200N @ Arch height 16
200N @ Arch Height 20.5
200N @ Arch Height 25
(C)
69
Figure 21: Varying arch height (only 3mm is shown, other orthotic thicknesses are
similar)
As the arch height changes so does the proportion of the arch deflection. The
higher the arch the less it deflects, so a person with a higher arch may be able to get
away with a thinner orthotic then for someone of the same weight with lower arches
depending on what their needs are.
Study II
Surrogate models are tools that can be quickly run in lieu of the full finite
element model in a more efficient manner. These models also allow for the ability to
minimize the error between a general template of an orthotic and specific patient’s
requirements of an orthotic. Specification of individual patient’s biomechanics, footfall
patterns, and activity level can be analyzed and added to the models quickly. Clinicians
can input specific factors into the surrogate models without having to run the full FEA
model, saving on ample computational time.
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
0 5 10 15 20 25 30
De
fle
ctio
n (
mm
)
Arch Height (mm)
Varying Arch Height (3mm Orthotic)
300lbs_3mm
140lbs_3mm
88lbs_3mm
70
We tested the performance of four algorithms on three design spaces. For each
model we looked at the percent difference between the Finite element model and the
surrogate model to assess the confidence of our results. The Kriging response surface
method seems to be the best out of the four methods. The average percent difference
for the Kriging method ranges between 0.23 and 0.30 %. The other methods range from
1.45 to 18% average difference. The time it takes to run the surrogate models were also
over 200% faster than the 6 minutes it took to run the basic FEA model. The small
values (below 0.5% difference) show that the Kriging method is accurate enough to be
used in lieu of running the full FE model and since the surrogate models run much
faster than the FE models, the Kriging method is more efficient without much loss of
accuracy. This method seems to be the best suited for predicting orthotic data and can
be used in the prescription process. These FEA models are in their most basic form,
once complexity is added to the models, so does the cost to run them. Adding patient
specification and non linear complex materials to create the full orthotic can multiply
the time to run by hours. It also may take several hundred FEA runs at each alteration;
the time savings for a specific CFO in office investigation for individual patients can be
much more feasible with the use of surrogate models.
Many of the other methods may work better with larger data sets, so it may be
possible to increase the accuracy of these other surrogate models by increasing the
model building data set. This may give the tools a better idea of the relationship
between the input and output information. The Kriging seems to work well with any
size data set due to the correlation function that performs a correction after each
71
iteration. The difference between the sample data and the Kriging function is calculated
and the function is adjusted according to the departure value.
When comparing the Kriging predictor model to Table 1, the values of the table
were slightly more conservative. As the table suggests that for a subject weighing
between 88lbs and 160lbs a recommended thickness for the orthotic is 4mm. The
clinical trial results demonstrated that for an 88 lb subject a 4mm orthotic will only
deflect approximately 1mm and for the 160lb subject the 4mm orthotic only deflected
1.82mm. This may be too stiff for both cases and cause the subject to have a thicker
orthotic unnecessarily. The Kriging predictor values are based on the clinical trial data
thereby demonstrating a more accurate guideline for clinicians to follow.
The Surrogate models are using the arch height and thickness of the orthotic to
create a relationship between the deflections of the arch, as can be seen from the
response surfaces in figures 14, 15 and 16. Although the arch affects the deflection as
was demonstrated by the first case study, the thickness is a much more determining
factor in the stiffness of the orthotic this is consistent with engineering principles.
Referring back to equation (2) the moment of inertia is the property of a beam that
predicts the resistance of beams to bending around the cross sectional plane. The
moment of inertia states that the stiffness is correlated to the thickness cubed.
Therefore the thickness is the main determinant in regards to stiffness of the orthotic.
Another draw to using surrogate models is the minimization of error through the
use of optimization. One example of the variability of use of these surrogate models for
optimal orthotic design is discussed in the case study below.
72
Case study: Force load distribution using surrogate modeling
In the previous study the validated finite element model of a custom foot
orthotic was introduced. This model simulated the stance phase loading simulation on
the surface of a custom foot orthotic using ANSYS WB. A simplified loading condition
was initially run, assuming the plantar pressure distributed uniformly across the entire
orthotic. Another set of simulations yielded a redistribution of the plantar pressure to
60/40, with 60% of the pressure in the heel and 40% in the front portion of the orthotic.
This set of simulations decreased the error between the uniformly distributed loaded
model and the clinical data. In another set of simulations the plantar pressure was
redistributed once more to 30% of the pressure in the heel and 70% in the front portion
of the model. For this study we will only be using the uniformly distributed load and
the 70/30 distribution for comparison.
The results plotted in Table 11 show a statistically significant amount of error
between the clinical data collected and the values obtained using FEA when using a
uniform loading condition. The resulting percent error and p-value for the 30/70
condition drastically improve. The percent error decreases by 15 to 24%, and the p-
value provides us with evidence that is not strong enough to reject the null hypothesis
that the Clinical and FEA data are equivalent. Overall, as the pressure distribution
across the orthotic better mimics the average plantar pressure of the human foot, the
smaller the percent error between the FEA model and clinical data and the greater the p-
value. Also note that up until this point percent difference has been used for the
comparison of values as neither value was regarded as the correct value the intent was
to evaluate the differences between them. For this study percent error is used because
73
the clinical data is considered the accepted value and we are comparing our results to
that value.
Table 11: Results from uniform and 30/70 simulations compared to clinical results.
The deformations are in negative z direction. The number in parenthesis under the
clinical deflection are standard deviation.
Loading Orthotic thickness
Clinical deflection accepted
FEA Model deflection measured
Percent Error
P-Value
t(mm) δ(mm) δ(mm) * **
Uniform 3 3.00 (0.25) 3.84 28 0.132
30/70 3 3.00 (0.25) 2.89 3.67 0.646
* [(Accepted-Measured)/Accepted]x100
** H0: δclinical = δFEA vs. HA: δclinical ≠ δFEA
The intent of this case study was to find the optimal plantar pressure distribution
for the finite element model to minimize the error between FEA model outputs and the
clinical data. This study was only run on the 3mm model of the medium weight subject.
Based on Cavanaugh’s 1987 plantar pressure distribution, the finite element model was
divided up into 10 anatomical regions depicted below in Figure 22. The weight
distribution regional mean values were used to portray a more accurate human plantar
pressure distribution. Cavanagh (Cavanagh, 1987) defined the 10 anatomical foot
regions using regional division method of footprints. Since the orthotic was cut off half
way between the metatarsal regions; the hallux, second toe and lateral toe will not be
considered. This eliminated region was only 3.6% of the total weight distribution so the
remaining regions equal 96.4%. The distribution of the weight along the pressure
regions according to Cavanagh (Cavanagh, 1987) is shown in Table 12.
74
Figure 22: The 10 anatomical regions that result from regional division (Cavanagh,
1987).
75
Table 12: Weight Distribution Regional Mean Values and SD (N=107) (Cavanaugh,
1987).
In order to attain an objective function for optimization, the CFO FE model was
altered in SolidWorks. The Geometry tool in ANSYS Workbench was then used to
imprint the 7 anatomical regions being investigated. In addition, ANSYS WB Design
Modeler and Mechanical were used to run various loading simulations to determine
which anatomical regions were most significant in arch deformation for the medium
subject. Finally, the Response Surface was created using the Kriging method in ANSYS
WB.
The original model was constructed with a division between the heel and forefoot
as shown in Figure 23(a). In order to create the 7 anatomical regions the division was
smoothed into a uniform surface and the anatomical divisions were imprinted onto the
top face and extruded down through all. The updated CFO FE model with anatomical
regions is shown in Figure 23(b).
76
Figure 23: (a) ANSYS workbench CFO model showing midline division, (b) ANSYS
workbench CFO model displaying imprinted faces for redistributed plantar pressure
distribution.
The load at each anatomical region was defined as a parameter that could be
altered in ANSYS to efficiently mimic the loading conditions outlined in Table 14. The
highlighted cells in Table 8 reflect the anatomical region that is maximized during the
particular loading condition. For example, someone who has a “medial heavy”plantar
pressure distribution focuses most of their weight on the medial side of their foot and
may possibly have flat arches.
Table 13: Pressure distribution (%) for various loading conditions with corresponding
deformation Mean (SD) of N=10
Region Average
Heel
heavy
Forefoot
heavy
Midfoot
heavy
Medial
heavy
Lateral
heavy
Medial heel 32.5 41.1 23.9 23.9 41.1 15.3
Lateral heel 28 36.2 50.6 20.6 20.6 36.2
Medial
Midfoot 1.4 2 1.2 2.5 2.5 0
Lateral
Midfoot 6.4 9 7.7 11.3 2.4 11.3
First MET 5.6 1.1 10.1 10.1 10.1 0.7
Second MET 8.4 4.6 12.2 7.3 12.2 12.2
LateralMETS 14.1 7.5 20.7 20.7 7.5 20.7
MEAN (SD) 2.96(0.074) 1.5(0.015) 4.47 (0.112) 4.33 (0.101) 5.36 (0.118) 1.09 (0.036)
(a)
(b
)
77
For each loading condition, the deformation of the highest point of the arch was
analyzed. At this location 10 sample data points were collected to determine the mean
deformations and the standard deviations presented in Table 13. From these results the
average plantar pressure distribution presented the most accurate deformation reading
when compared to the clinical data (3mm deflection). Therefore, for the remainder of
this study, an average plantar pressure distribution (Cavanaugh, 1987) was assumed for
the medium subject.
For the generated surrogate models, the data used were split up into two different
data layouts, the first (will be referred to as MH) using the Lateral heel (LH) and Lateral
metatarsals (LM) as input design variables, the calculated design variables being the
Medial heel (MH), and all other anatomical regions were constant. The second data
layout (will be referred to LM) was setting the MH and LH regions as input design
variables, LM’s region was the calculated design variable, and all other regions were held
constant.
A face centered central composite design was used to generate the design of
experiments. Each scenario provided 17 design points that were used to construct the
design surface using the Kriging methodology to predict the relationship between the
design variables and the response variable.
The three types of optimization used in ANSYS Workbench Goal Driven
Optimization are Screening, MORG, and NLPQL. The Screening approach is a non-
iterative direct sampling method by a quasi-random number generator based on the
Hammersley algorithm. The MOGA approach is an iterative Multi-Objective Genetic
Algorithm, which can optimize problems with continuous input parameters. NLPQL is a
78
gradient based single objective optimizer that is based on quasi-Newton methods
(ANSYS, 2009). Three different sets of initial points were used to ensure a global
optimal solution was achieved.
The Screening Optimization Method uses 10000 initial samples with a Constraint
Handing “As Goals” and a generated sample set of 1. This allows for the Objective in the
Optimization study of the deformation output parameter to be assigned as “Seek Target”
with a value of -3mm. The importance of this main objective is set to “higher”. This
technique results in three local optimal solution candidates as displayed in Table 14.
The MOGA Optimization Method also uses 10000 initial samples with a
Constraint Handing “As Goals” and a generated sample set of 1. In addition, there are
100 samples per iteration, a maximum allowable Pareto Percentage of 70 and a maximum
number of iteration of 50. The deformation response variable objective is set to “Seek
Target” with a value of -3mm and a higher importance. This technique also results in
three local optimal solution candidates as shown in Table 14.
Table 14: Optimal Design Variable Pressure Distributions and corresponding
deformation values from various Goal Driven Optimization Techniques
Optimization
Technique Deformation
(mm) Lateralheel Medialheel LateralMETS
Screening 1 -3.0002 0.14642 0.48862 0.11096
Screening 2 -2.999 0.14262 0.4389 0.16448
Screening 3 -3.0066 0.14137 0.36903 0.2356
MOGA 1 -3.0046 0.19372 0.32573 0.22655
MOGA 2 -2.9925 0.13672 0.33723 0.27204
MOGA 3 -2.9531 0.21571 0.30536 0.22493
NLPQL -3 0.28695 0.29992 0.15912
The NLPQL Optimization Method has an allowable convergence percentage of
1E-10, a maximum number of iterations of 50, and a generated sample set of 1. The
79
Constraint Handling is set to “As Goals” allowing the deformation objective to be set to
“Seek Target” with a value of -3mm and a higher importance. This particular technique
results in one global optimal solution as highlighted in Table 14.
The objective of this study was to show that as the plantar pressure distribution is
more specifically defined to an individual the closer the FEA deflection values match up
with the clinical deflection values as laid out in Table 13. The optimization feature was
utilized to find the optimal distribution by minimizing the error to zero and increasing the
P-value to 1 accepting the null hypothesis that the clinical data and FEA data are
equivalent as can be seen in Table 15.
Table 15: Uniform, 30/70 and Optimal Pressure FEA results compared to clinical data
(the deformations are in the negative z-direction)
Loading Orthotic
thickness Clinical deflection
accepted
FEA Model
deflection
measured
Percent
Error P-Value
t(mm) δ(mm) δ(mm) * **
Uniform 3.00 3.00 (0.25) 3.84 28.00 0.13
30/70 3.00 3.00 (0.25) 2.89 3.67 0.65
Optimal 3.00 3.00 (0.25) 3.00 0.00 1.00
* [(Accepted-Measured)/Accepted]x100
** H0: δclinical = δFEA vs. HA: δclinical ≠ δFEA
The optimal pressure distribution (Table 16) found in the optimization for the
medium subject can be considered a variation of Cavanaugh’s average distribution.
80
Table 16: Optimal Plantar pressure Distribution for medium subject.
Anatomic Region Pressure Distribution (%)
Medium Heel 29.99
Lateral Heel 28.7
Medial Mid-foot 1.4
Lateral Mid-foot 6.4
First Metatarsal 5.6
Second Metatarsal 8.4
Lateral Metatarsals 15.91
TOTAL 96.4
81
CHAPTER 6
SUMMARY & FUTURE WORK
Summary
These studies clearly indicate that modeling with FEA techniques offers a
consistent, accurate and reliable alternative to clinical studies, which contain many other
external uncontrollable factors, establishing a foundation for a methodical approach to
engineering modeling of orthotics. This approach offers a powerful framework to mimic
experimental and clinical study data and therefore can be a viable and valuable tool in the
custom design of orthotics based on individual’s unique needs and foot characteristics
without requiring extensive and expensive trial and error ad hoc approaches.
Each of the four surrogate modeling methods uses different algorithms to create a
surface, which describes the relationship between the input variables and the output data.
The Kriging method seems to have the ability to describe the custom foot orthotic
relationship the best for the purposes of this research. These Kriging surrogate models
are extremely efficient and accurate enough to be used in the custom foot orthotic
prescription process. The other methods are not advisable to be used for these purposes
and may be better suited for other relationships or situations with larger data sets.
Furthermore, these engineering models are easy and straightforward to modify for
specific use in practical situations. Such models can be flexible and adaptive to include
other design considerations, such as the activity factor, and foot deformities as seen in the
second case study. With the new and enhanced modeling capabilities, the CFO
prescriber can have the ability to be able to design and develop the best-fit CFO with the
optimal design characteristics for individual patients. Such a model could also enable
82
visual inspection of the impact of small changes in the input conditions on the overall
performance of the CFO.
Future work
These models open up a whole new world of studies to advance the knowledge
base on CFO’s. With these engineering models in conjunction with clinical trials many
developments can be made. Modeling is a tool that can greatly improve the methods that
are used to determine the effectiveness of orthotics. Some of the initial future research
may consist of: 1) running a more comprehensive clinical trial to gather more data on
specific groups to increase the accuracy of the FEA model. 2) Further investigate the
varying load distribution from the case study. Expand to include other subjects and
groups of people including obese populations which have been shown to carry their
weight very differently than “normal” populations. Also include a validation of the case
study by collecting the pressure distribution from the “medium” subject. 3) A clinical
study may also be run with only specific subject such as an “obese” group of subjects
collecting the pressure distribution between the foot and orthotic. This may be helpful
for both the first and second studies on this list. 4) Develop the Kriging surrogate models
in MATLAB in order to have use of equations. This may allow for more variability and
specification with the models including which correlation models to use. This may also
allow for the possibility of replacing the empirical table (Table 1) with the response
surface to include reference points for clinicians to easily reference. 5) Layer the FEA
model with soft nonlinear materials used in a full CFO. Including the soft materials will
create an even more accurate representation of the CFO behavior for future research and
design. 6) Investigate other stance phases such as heel strike and toe off.
83
APPENDIX
CLINICAL TRIAL DOCUMENTATION
84
ADULT INFORMED CONSENT DOCUMENT
University of Massachusetts
Amherst, MA 01003
Title: The validation of the design model of custom foot orthotics.
Principal Investigators: Lieselle Trinidad, MS; Sundar Krishnamurty, PhD; Ryan
Chang, MS; and Joseph Hamill, PhD.
Your written informed consent is required before you can participate in this project. Please read this document carefully and
then sign your name on the last page if you agree to participate. This document is in accordance with the Assurance of Compliance with the Office of Human Research Protection Regulations as approved by the Faculty Senate of the University of
Massachusetts.
Purpose: The purpose of this study is to validate the arch deflection of a design model
(FEA model) of a custom foot orthotic.
Eligibility: To participate in this study, you must be 10 to 60 years of age. You do not,
and have no history of: severe structural foot abnormality, arthritis, neurological
disorders, myopathies, cardiovascular disorder in the foot, foot infections and tumors.
Procedures:
This will be carried out at the University of Massachusetts, Biomechanics Laboratory
(Totman Building Room 23). You will complete a Modified Physical Activity Readiness
Questionnaire to determine your overall ability to participate in exercise. There will be a
total of two session to complete this study.
Session 1: We will measure your weight and end by making a cast of your foot from
which we will build a custom foot orthotic. This session will take approximately 30
minutes.
Session 2 (Motion Analysis Session): This will be carried out at the University of
Massachusetts, Biomechanics Laboratory. You will be asked to perform four tasks while
wearing the custom foot orthotics: 1) sit, 2) stand on one foot, 3) walk, and 4) run.
Before you begin any tasks, the custom foot orthotic will be taped to your bare skin and
reflective markers will be placed at three locations on the custom foot orthotic. The
movements of the reflective markers will be captured by cameras as you walk into their
recording area. After the orthotic and reflective markers are attached to your foot you will
sit on a chair with your foot placed on the force platform holding still for approximately
10 seconds, this is done to get a baseline reading of the arch height of the orthotic. This
task will be repeated 2-3 times. For the second task, you will be asked to stand on the
force platform on the one foot that has the orthotic attached to it. You will use a stick to
help you balance, and you will need to hold this position for about 10 seconds. That task
will be repeated 10 times. For the third and forth task you will be asked to walk then run
from one end of the laboratory to the other across the force platform making sure to place
85
the foot with the orthotic attached to it on the force platform. This will take
approximately 10 seconds and will be repeated 10 times for each activity. You will be
provided with rest periods as needed. At the end of the procedure, all markers will be
removed. Once you have completed all four tasks with the right foot we will ask you to
repeat the same sequence for the left foot. In total, this session should take approximately
90 minutes.
Possible Risks and Discomforts: The following risks and discomforts are associated
with the procedures described above.
Motion Analysis Session. During any type of exercise, there are slight possibilities of
health risks such as temporary fatigue and muscle soreness.
Confidentiality: Your identity and records will be kept confidential. While results from
this study will be shared with other researchers, no individual identities will be used in
any reports or publications resulting from this study.
In Case of Injury: In the unlikely event of injury resulting directly from participation in
this study, we will do everything we can to assist you in seeking medical treatment. The
University of Massachusetts will not provide compensation for medical treatment you
obtain.
Benefits: You will receive no direct benefit from participating in this study. Any
information that is obtained from this study will be made available to your physician,
upon request. The purpose of these studies is to provide the investigators with
information that will help us validate a design model of custom foot orthoses. This
information ultimately may have a positive impact on the research and development of
custom foot orthoses.
Costs and Reimbursement: No costs will be charged to you if you participate in this
study. You will receive one pair of foot orthoses after completing the study.
Withdrawal of Participation: Participation in this research is voluntary. You have the
right to withdraw from this study at any time.
Information: You are encouraged to ask questions about the study. The investigators
will attempt to answer all of your questions to the best of their knowledge. The
investigators fully intend to conduct the study with your best interest, safety and comfort
in mind. Please address any questions regarding the study Dr. Sundar Krishnamurty,
Ph.D. at [email protected], or to Lieselle Trinidad, M.S. (716) 310-7854. If you
would like to speak with someone not directly involved in the research study, you may
contact the Human Research Protection Office at the University of Massachusetts via
email at [email protected]; telephone (413) 545-3428; or mail at the Human
Research Protection Office, Research Administration Building, University of
Massachusetts Amherst, 70 Butterfield Terrace, Amherst, MA 01003-9242.
86
Participant’s Name Address
Signature Phone Number Date
______________________________
Investigator Signature
Department of Mechanical Engineering
87
MODIFIED PHYSICAL ACTIVITY READINESS QUESTIONAIRE
Initial Screening: Interview Date (MM/DD/YY):
______/______/______
Last Name _______________________ First Name ___________________________
Phone # Phone #
Age (yrs) ______________ DOB: _____________________ Gender:
Female / Male
Height: _____ Feet, _____ Inches or ________ cm
Weight: ________________lbs or _________ kg
General health status _____________
Are you on medication?
Yes No Do you or have a significant past medical history? (eg. surgery, hospitalization )
__________________________________________________________
Yes No Any injuries in past the would affect walking?
Yes No Do you have physical limitations?
Yes No Do you have any heart problems?
Yes No Do you have diabetes?
Yes No Do you have arthritis?
Yes No Do you have neuropathies?
Yes No Circulations disorders? e.g. swelling of discoloration of your feet?
How did you hear about the study?__________________________________
88
Modified Physical Activity Readiness Questionnaire
1. Yes No Has your doctor ever said you had heart trouble or a heart murmur?
2. Yes No Do you ever suffer pains in your chest?
3. Yes No Do you ever feel faint or have spells of severe dizziness, passed
out, palpitations or rapid heart beat?
4. Yes No Has the doctor ever told you that your blood pressure was too high? (systolic >
160 mm Hg or diastolic > 90 mm Hg on at least 2 separate occasions)
5. Yes No Do you smoke cigarettes?
6. Yes No Do you have diabetes?
7. Yes No Do you have a family history of coronary or other atherosclerotic
disease in parents or siblings prior to age 55?
8. Yes No Has your serum cholesterol ever been elevated?
9. Yes No Is there any physical reason not mentioned here why you should
not follow an activity program even if you wanted to?
Below please provide an explanation for any of the questions to which you answered
YES.
________________________________________________________________________
________________________________________________________________________
____________________________________________________
Body Measurements
Height: _____ Feet, _____ Inches or ________ cm
Weight: ________________lbs or _________ kg
89
PARENTS INFORMED CONSENT DOCUMENT
University of Massachusetts
Amherst, MA 01003
Title: The validation of the design model of custom foot orthotics.
Principal Investigators: Lieselle Trinidad, MS; Sundar Krishnamurty, PhD; Ryan
Chang, MS; and Joseph Hamill, PhD.
Your written informed consent is required before your child can participate in this project. Please read this document carefully
and then sign your name on the last page if you agree to allow your child to participate. This document is in accordance with the Assurance of Compliance with the Office of Human Research Protection Regulations as approved by the Faculty Senate of the
University of Massachusetts.
Purpose: The purpose of this study is to validate the arch deflection of a design model
(FEA model) of a custom foot orthotic.
Eligibility: To participate in this study, participants must be 10 to 60 years of age.
Participant does not, and has no history of: severe structural foot abnormality, arthritis,
neurological disorders, myopathies, cardiovascular disorder in the foot, foot infections
and tumors.
Procedures:
This will be carried out at the University of Massachusetts, Biomechanics Laboratory
(Totman Building Room 23). Your child will complete a Modified Physical Activity
Readiness Questionnaire to determine his/her overall ability to participate in exercise.
There will be a total of two session to complete this study.
Session 1: We will measure your child’s weight and end by making a cast of his/her foot
from which we will build a custom foot orthotic. This session will take approximately 30
minutes.
Session 2 (Motion Analysis Session): This will be carried out at the University of
Massachusetts, Biomechanics Laboratory. Your child will be asked to perform four tasks
while wearing the custom foot orthotics: 1) sit, 2) stand on one foot, 3) walk, and 4) run.
Before your child begins any tasks, the custom foot orthotic will be taped to his/her bare
skin and reflective markers will be placed at three locations on the custom foot orthotic.
The movements of the reflective markers will be captured by cameras as your child walks
into their recording area. After the orthotic and reflective markers are attached to your
child’s foot your child will sit on a chair with his/her foot placed on the force platform
holding still for approximately 10 seconds, this is done to get a baseline reading of the
arch height of the orthotic. This task will be repeated 2-3 times. For the second task,
your child will be asked to stand on the force platform on the one foot that has the
orthotic attached to it. Your child will use a stick to help him/her balance, and he/she
will need to hold this position for about 10 seconds. That task will be repeated 10 times.
90
For the third and forth task your child will be asked to walk then run from one end of the
laboratory to the other across the force platform making sure to place the foot with the
orthotic attached to it on the force platform. This will take approximately 10 seconds and
will be repeated 10 times for each activity. Your child will be provided with rest periods
as needed. At the end of the procedure, all markers will be removed. Once your child
has completed all four tasks with the right foot we will ask him/her to repeat the same
sequence for the left foot. In total, this session should take approximately 90 minutes.
Possible Risks and Discomforts: The following risks and discomforts are associated
with the procedures described above.
Motion Analysis Session. During any type of exercise, there are slight possibilities of
health risks such as temporary fatigue and muscle soreness.
Confidentiality: Your child’s identity and records will be kept confidential. While
results from this study will be shared with other researchers, no individual identities will
be used in any reports or publications resulting from this study.
In Case of Injury: In the unlikely event of injury resulting directly from participation in
this study, we will do everything we can to assist your child in seeking medical treatment.
The University of Massachusetts will not provide compensation for medical treatment
your child obtains.
Benefits: Your child will receive no direct benefit from participating in this study. Any
information that is obtained from this study will be made available to your child’s
physician, upon request. The purpose of these studies is to provide the investigators with
information that will help us validate a design model of custom foot orthoses. This
information ultimately may have a positive impact on the research and development of
custom foot orthoses.
Costs and Reimbursement: No costs will be charged to you or your child if you
participate in this study. Your child will receive no reimbursement for participation in
this study.
Withdrawal of Participation: Participation in this research is voluntary. Your child has
the right to withdraw from this study at any time.
Information: You and your child are encouraged to ask questions about the study. The
investigators will attempt to answer all of your questions to the best of their knowledge.
The investigators fully intend to conduct the study with your child’s best interest, safety
and comfort in mind. Please address any questions regarding the study Dr. Sundar
Krishnamurty, Ph.D. at [email protected], or to Lieselle Trinidad, M.S. (716)
310-7854. If you would like to speak with someone not directly involved in the research
study, you may contact the Human Research Protection Office at the University of
Massachusetts via email at [email protected]; telephone (413) 545-3428; or
91
mail at the Human Research Protection Office, Research Administration Building,
University of Massachusetts Amherst, 70 Butterfield Terrace, Amherst, MA 01003-9242.
Participant’s Name Address
Parent/Guardian Signature Phone Number Date
______________________________
Investigator Signature
Department of Mechanical Engineering
92
BIBLIOGRAPHY
Abu-Hasaballah, Khamis S., Michael D. Nowak and Paul D. Cooper. "Enhanced
Solid Ankle-Foot Orthosis Design: Real-Time Contact Pressures Evaluation and
Finite Element Analysis." Proceedings of the 1997 ASME International
Mechanical Engineering Congress and Exposition, Nov 16-21 1997.
Actis, Ricardo L., Ventura, Liliana B., Smith, Kirk E., Commean, Paul K., Lott,
Donovan J., Pilgram, Thomas K. and Mueller, Michael J.,“Numerical simulation
of the plantar pressure distribution in the diabetic foot during the push-off stance”
Medical and Biological Engineering and Computing 44.8 (2006): 653-663
Arnold, John B., et al. "The Impact of Increasing Body Mass Peak and Mean in
Asymptomatic Adults Subjects during Walking." Diabetic Foot & Ankle 1 (2010).
Barani, Zohreh, Haghpanahi, Mohammad, and Katoozian, Hamid “Three dimensional
stress analysis of diabetic insole: A finite element approach” Technology and Health
Care 13.3 (2005): 185-192.
Bar-Yam, Yaneer. "Dynamics of Complex Systems." Westview Press, 2003.
Batavia, Mitchell. Clinical Research for Health Professionals. 1st ed. Butterworth-
Heinemann, 2000.
Bates, B.T., Osternig, L.R., Mason, B., James, L.S.,"Foot Orthotic Devices to Modify
Selected Aspects of Lower Extremity Mechanics." The American Journal of Sports
Medicine 7.6 (1979): 338-42.
Bates, Barry T., “Single-Subject methodology: an alternative approach” Medicine &
Science in Sports & Exercise 28.5 (1996): 631-638
Biewener, Andrew and Robers, Thomas J., “Muscle and Tendon Contributions to
Force, Work, and Elastic Energy Savings: A Comparative Perspective”Exercise
and Sport Sciences Reviews 28.3 (2000): 99-107.
Birtane, Murat, and Tuna Hakan. "The Evaluation of Plantar Pressure Distribution in
Obese and Non-Obese Adults." Clinical Biomechanics 19 (2004): 1055-1059.
Bradley, Nuran. THE RESPONSE SURFACE METHODOLOGY. Indiana University
South Bend, 2007 South Bend, Indiana.
Breiman, Leo, et al. Classification and Regression Trees. Monterey, CA: Wadsworth &
Brooks/Cole Advanced Books & Software, 1984.
Board on Manufacturing and Engineering Design, National Research Council, (2001),
Theoretical Foundations for Decision Making in Engineering Design, National
Academy Press, http://www.nap.edu/catalog/10566.html, access date: 4/11/2008.
93
Box, G. E. P., Hunter, W. G., and Hunter, J. S., (1978), Statistics for Experimenters:
An Introduction to Design, Data Analysis and Model Building, ISBN:
0471093157, John Wiley and Sons, Inc., New York, NY.
Cavanagh, P. R., M. M. Rodgers, and A. Iiboshi. "Pressure Distribution Under Symptom-
Free Feet during Barefoot Standing." Foot Ankle 7 (1987): 262-76.
Chen, Weng-Pin, Chia-Wei Ju, and Fuk-Tan Tang. "Effects of Total Contact Insoles
on the Plantar Stress Redistribution: A Finite Element Analysis." Clinical
Biomechanics 18.6 (2003): 17-24.
Cheung, Jason Tak-Man, Zhang, Ming, “Parametric design of pressure-relieving foot
orthosis using statistics-based finite element method” Archives of Physical Medicine
and Rehabilitation 86 (2005), 353-358
Cheung, Jason Tak-Man, Zhang, Ming, “Parametric design of pressure-relieving foot
orthosis using statistics-based finite element method” Medical Engineering & Physics
30 (2008): 269–277
Cheung, Jason Tak-Man, Zhang, Ming, “Finite Element Modeling of the Human Foot
and Footwear” 2006 ABAQUS Users’ Conference
Chu, D, (2004)“The GLOBEC Kriging Software Package – EasyKrig3.0,” Woods
Hole Oceanographic Institution.
Chu, T. M., and N. P. Reddy. "Stress Distribution in the Ankle-Foot Orthosis used to
Correct Pathological Gait." Journal of rehabilitation research and development
32.4 (1995): 349-60.
Dierks, Tracy A., Manal, Kurt T., Hamill, Joseph, Davis, Irene S., “Proximal and
Distal Influences on Hip and Knee Kinematics in Runners with Patellofemoral
Pain During a Prolonged Run” Journal of Orthopaedic & Sports Physical Therapy
38.8 (2008): 448-456.
Doorenbosch, Caroline A.M., Van Ingen Schenau, Gerrit Jan, “The Role of mono-and
bi-articular muscles during contact control leg tasks in man” Human Movement
Science 14 (1995): 279-300.
Eng, J., and M. R. Pierrynowski. "The Effect of Soft Foot Orthotics on Three-
Dimensional Lower Limb Kinematics during Walking and Running." Physical
Therapy 74.9 (1994): 836-844.
Eng, J., and M. R. Pierrynowski. "Evaluation of Soft Foot Orthotics in the Treatment of
Patella Femoral Pain Syndrome." Physical Therapy 73.2 (1993): 62-68.
Erdemir, Ahmet, Saucermanb, Jeffrey J., Lemmonc, David, Loppnowd, Bryan, Brie,
Tursoe, Ulbrechtf, Jan S. and Cavanagh, Peter Re, “Local Plantar pressure relief in
therapeutic footwear: design guidelines from finite element models” Journal of
Biomechanics 38.9 (2005): 1798-1806
94
Evans, R. Scott. "Common Statistical Concerns in Clinical Trials." J exp stroke transl
med 3.1 (2010): 1-7.
Faustini, Mario C., Neptune, Richard R., Crawford, Richard H., Stanhope, Steven J.,
“Manufacture of Passive Dynamic nkle-Foot Orthoses Using Selective Laser
Sintering” IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING 55.2
(2008): 784-790.
Feldman, Neil J., and Roberts, Michael, “Orthotics: Worth the Cost?” F.I.R.M. racing
newsletter May 2010
Ferber, Reed, Davis, Irene McClay, Williams, Dorsey s. III, “Effect of foot orthotics on
rearfoot and tibia joint coupling patterns and variability” Journal of Biomechanics 38
(2005): 477-483.
Finney, Robert H., and Alok Kumar. "Development of Material Constants for
Nonlinear Finite-Element Analysis." Rubber Chemistry and Technology 61.5
(1988): 879-91.
Giunta, A. A., and Watson, L. T., (1998), "A Comparison of Approximation
Modeling Techniques: Polynomial versus Interpolating Models," AIAA 98-4758.
Hamill, Joseph, Van Emmerik, Richard E.A., Heiderscheit, Bryan C., Li, Li, “A
Dynamic Systems Approach to Lower extremity Running Injuries” Clinical
Biomechanics 14 (1999): 297-308.
Hardin, E.C., Hamill, J., “The Influence of Midsole Cushioning on Mechanical and
Hematological Responses During a Prolonged Downhill Run” Research Quarterly
for Exercise and Sport 73.2 (2002): 125-133.
Hazelrigg, G.A., (1999), “On the Role and Use of Mathematical Models in
EngineeringDesign,” ASME Journal of Mechanical Design, 121, 336-341.
Hills, AP, et al. "Plantar Pressure Differences between Obese and Non-Obese Adults: A
Biomechanical Analysis." International Journal of obesity 24 (2001): 1674-1679.
Juvinall, Robert C., and Kurt M. Marshek. Fundamentals of Machine Component Design.
3rd ed. John Wiley and Sons, 2000.
Kleijnen, J. P. C., and Beers, W. C. M. v., (2003), "Kriging for Interpolation in
Random Simulation," Journal of the Operational Research Society, 54, 255-262.
Kleijnen, J. P. C., and Sargent, R. G., (2000), "A Methodology for Fitting and
Validating Metamodels in Simulation," European Journal of Operational
Research, 120(1), 14-29.
Kutner, Nachtsheim, and Li Neter. Applied Linear Statistical Models. 5th ed. McGraw-
Hill/Irwin, 2004.
95
Lam, P. C., et al. "FINITE ELEMENT ANALYSIS OF 'DROP-FOOT'
ORTHOSES." 1985 Advances in Bioengineering. Presented at the Winter Annual
Meeting of the American Society of Mechanical Engineers 17 Nov 1985; Miami,
FL, USA
Lawrence, Jeanette. "Introduction to Neural Networks." (1994).
Leardini, A., Benedetti, F., Catani, F., Simoncini, L., Giannini, S., “An anatomically
based protocol for the description of foot segment kinematics during gait”
Clinical Biomechanics 14 (1999): 528-536.
Lemmon, David R., et al. "Finite Element Model of the Forefoot Interface with
Accommodative Footwear Orthoses." Proceedings of the 1995 ASME
International Mechanical Engineering Congress and Exposition, Nov 12-17 1995.
Lemmon, David, et al. "Effect of Insoles in Therapeutic Footwear - a Finite Element
Approach." Journal of Biomechanics 30.6 (1997): 615-20.
Leone, D., S. Diemente and M. Lopez-Isa. "Structural Stability Prediction for
Thermoplastic Ankle-Foot Orthoses." Proceedings of the 1991 IEEE 17th Annual
Northeast Bioengineering Conference, Apr 4-5 1991.
MacLean C, Davis IM, Hamill J. "Influence of a Custom Foot Orthotic Intervention
on Lower Extremity Dynamics in Healthy Runners." Clin Biomech (Bristol, Avon
6 (2006): 623,624, 625, 626, 627, 628, 629, 630.
MacLean, C., I. M. Davis, and J. Hamill. "Influence of a Custom Foot Orthotic
Intervention on Lower Extremity Dynamics in Healthy Runners." Clinical
biomechanics (Bristol, Avon) 21.6 (2006): 623-30.
Martin, J. D., and Simpson, T. W., (2005), "Use of Kriging Models to Approximate
Deterministic Computer Models," AIAA Journal, 43(4), 853-863.
McCune, B., and M. J. Mefford. "HyperNiche. Nonparametric Multiplicative Habitat
Modeling." MjM Software (2004).
McCune, B. "Non-Parametric Habitat Models with Automatic Interactions." Journal of
Vegetation Science 17: 819–830.
Menz, Hylton B., and Meg E. Morris. "Clinical Determinants of Plantar Forces and
Pressures during Walking in Older People." Gait & Posture 24 (2006): 229-236.
Montgomery, Douglas C. Design and Analysis of Experiments. 4th Edition ed. New
York: John Wiley and Sons, 1997.
Morin, Jean-Benoit, Dalleau, Georges, Kyrolainen, Heikki, Jeannin, Thibault, Belli,
Alain, “A Simpple Method for Measuring Stiffness During Running” Journal of
Applied Biomechanics 21 (2005): 167-180.
96
Morton, DJ. "The Human Foot: Its Evolution, Physiology, and Functional Disorders."
(1935).
Mundermann, A., et al. "Foot Orthotics Affect Lower Extremity Kinematics and
Kinetics during Running." Clinical biomechanics (Bristol, Avon) 18.3 (2003):
254-62.
Myers, Raymond H., and Douglas C. Montgomery. Response Surface Methodology. New
York: John Wiley and Sons, 1995.
Nawoczenski, D. A., T. M. Cook, and C. L. Saltzman. "The Effect of Foot Orthotics
on Three-Dimensional Kinematics of the Leg and Rearfoot during Running." The
Journal of orthopaedic and sports physical therapy 21.6 (1995): 317-27.
Prescription Foot Orthotics Laboratory Association, Access date January 2008
www.PFOLA.org/technicaltopics
Richie,Douglas Jr., “Inside Insights on the Evolution of Orthotic Therapy”
Podiatrytoday.com 19.2 (2006)
Root, M. L. "Development of the Functional Orthosis." Clinics in podiatric medicine
and surgery 11.2 (1994): 183-210.
Sacks, J., Welch, W. J., Mitchell, T. J., and Wynn, H. P. (1989), "Design and
Analysis of Computer Experiments," Statistical Science, 4(4), 409-435.
Sargent, R. G., (2004), "Validation and Verification of Simulation," proceedings of
the 2004 Winter Simulation Conference, Ingalls, R .G., Rossetti, M. D., Smith, J.
S., and Peters, B. A.,eds. 17-28.
Saxena, A., and J. Haddad. "The Effect of Foot Orthosis on Patella Femoral Pain
Syndrome." Journal of American Podiatric medical association 93.4 (2003): 264-271.
Segal, M. R. "Tree-Structured Methods for Longitudinal Data." Journal of the American
Statistical Association 87.418 (1992).
Shao, T., and Krishnamurty, S., (2006), "Modeling Implications in Simulation-Based
Design of Stents," ASME 2006 International Design Engineering Technical
Conferences &Computers and Information in Engineering Conference,
Philadelphia, Pennsylvania, DETC2006-99493.
Shao, T., (2007), “Towards a structured approach to simulation-based engineering
design under uncertainty,” (Doctoral Dissertation, University of Massachusetts
Amherst, September 2007).
Shorten, M.R., and Himmelsbach, J.A., “Shock Attenuation of Sports Surfaces” The
Engineering of Sport IV:Proceedings of the 4th
International Conference of The
Engineering of Sport, Kyoto, 3-6 September 2002
Simpson, T. W., Peplinski, J., Koch, P. N.; and Allen, J. K., (1997), “On the Use of
Statistics in Design and the Implications for Deterministic Computer
97
Experiments,” ASME 1997 Design Engineering Technical Conferences,
Sacramento, CA, DETC97/DTM-3881.
Simpson, T. W., Mauery, T. M., Korte, J. J., and Mistree, F., (1998), "Comparison of
Response Surface and Kriging Models for Multidisplinary Design Optimization,"
AIAA 98-4755.
Simpson, T. W., Peplinski, J., Koch, P. N., and Allen, J. K., (2001), "Metamodels for
Computer-Based Engineering Design: Survey and Recommendations,"
Engineering with Computers, 17(2), 129-150.
Simpson, T. W., et al. "Approximation Methods in Multidisciplinary Analysis and
Optimization: A Panel Discussion." Structural and Multidisciplinary Optimization 27.5
(2004): 302-313.
Smith, L. S., et al. "The Effects of Soft and Semi-Rigid Orthoses upon Rearfoot
Movement in Running." Journal of the American Podiatric Medical Association
76.4 (1986): 227-33.
Smith, Murray. Neural Networks for Statistical Modeling. Van Nostrand Reinhold,
1993.
Stackhouse, C. L., I. M. Davis, and J. Hamill. "Orthotic Intervention in Forefoot and
Rearfoot Strike Running Patterns." Clinical biomechanics (Bristol, Avon) 19.1
(2004): 64-70.
Steindler, Arthur. "Kinesiology of the Human Body." (1955).
Tenbroek, Trampas. Doctoral Dissertation. University of Massachusetts Amherst, 2011
Amherst, ma.
Tessem, Siri, Nora Hagstrom, and Bjorg Fallang. "Weight Distribution in Standing and
Sitting Positions, and Weight Transfer during Reaching Tasks, in Seated Stroke
Subjects and Healthy Subjects." Physiotherapy Research International 12 (2007): 82-
94.
Theodoridis, Sergios, and Konstantinos Koutroumbas. Pattern Recognition. 4th ed.
Academic Press, 2009.
Vardeman, Stephen B., and Marcus J. Jobe. Basic Engineering Data Collection and
Analysis. 1st ed. Duxbury Press, 2000.
Vaughan, Christopher L., “Theories of bipedal walking: an odyssey” Journal of
Biomechanics 36 (2003): 513-523.
98
Verseput, Richard. "Digging into DOE: Selecting the right central composite Design for
response surface methodology applications." 2000.
<http://www.qualitydigest.com/june01/html/doe.html>.
Williams, D. S.,3rd, I. McClay Davis, and S. P. Baitch. "Effect of Inverted Orthoses
on Lower-Extremity Mechanics in Runners." Medicine and science in sports and
exercise 35.12 (2003): 2060-8.